A CFL with integrated ballast is a complex piece of electronics, made out of glass, steel, aluminium, solder, copper wire, silicon, phosphors, fibreglass, various plastics, ferrites or ceramics, epoxy resins  and flame retardants. Different parts (e.g. ballasts) are often manufacturered in different places, sometimes in different countries, and then sent to the main factory for final assembly. See ballast interior here: CFL autopsy
Some have attempted to make ‘cradle-to-grave’ estimates on CFL energy use, mercury emissions etc., so-called Life Cycle Assessments, but this is no easy task and results may vary depending on how many factors are included into the calculation, and what you mean by ‘cradle’ and ‘grave’.
For CFL proponents, ‘cradle’ means when the parts get assembled at the factory, and ‘grave’ means when they’re returned to a recycling facility or end up in landfill. For realists, ‘cradle’ means when the mercury, phosphors and rare earths are mined out of the ground and ‘grave’ when the mercury ends up either in a new lamp, or via nature and the food chain, in us. I’m sure no CFL proponent wants to include the costs for brain damaged babies and lowered general health and mental function of future generations through slow mercury poisoning of the entire population.
Greenpeace LCA Study
But even without adding the last to the calculation, realistic assessments like the one done by Klaus Stanjek on behalf of Greenpeace Hamburg, show that CFLs’ complicated construction may require 10 to 40 times more energy – and emissions – to produce than it takes to manufacture an incandescent bulb.  Even if they outlast 5-10 bulbs, and use less electricity in the use phase, they still seem to require more energy during their whole life cycle.
Compare with how simple it is to manufacture an incandescent bulb:
“It can also be argued that the incandescent bulb is quite environmentally friendly. Unlike higher technology lamps, the simple filament bulb does not require rare earth gases and phosphors, leaches no mercury, and requires no proprietary manufacturing patents. The incandescent light bulb is produced worldwide, and is often a local product, which requires less packaging and less fuel for transport from low-wage factories to high-profit markets.” – Jeff Miller, President-elect IALD, Director Pivotal Lighting 
Danish LCA study (mercury & coal)
For those who still believe that incandescent bulbs “cause more mercury emissions via coal plants”, please understand that it is nothing but a cheap PR trick which seems to originate from the pro-CFL/anti-lightbulb lobby organisation IAEEL 1993, and based alternately on:
I. U.S. conditions in which, at that time, 59% of electricity production came from coal.  June 2008 it was 48,5% and decreasing. 
II. A Danish ‘study’ (= calculation excercise) from 1991  in which a 60W (730 lm) 1000h incandescent (GLS) was compared with a 15W (900 lm) 8000h CFL, the latter assumed to contain 0.69 mg mercury, while electricity production from coal was assumed at 95%, as was the case in Denmark at that time – the highest in Europe! . Based on these assumptions, CFLs were estimated to emit 1.69 mg mercury per million lumen-hour during production, operation and crapping phase, and incandescents 4.86 mg. However, these figures were seriously flawed then, and are even more so today:
a. “0.69 mg mercury” in CFLs is seems like a random fantasy figure, especially back in 1991! In 1993, IAEEL estimated CFLs to contain an average of 5 mg.  Eu consultants VITO consider 4 mg to be a realistic average now.  (Both are extremely pro-CFL and are not likely to exaggerate.)
b. According to EuroStat, the EU share of coal used in electricity production was 39% in 1991 and has since decreased to 29% in 2006 (though varying widely between different countries). 
Correcting for a and b (while still assuming the 15W CFL to give as much light as a 60W GLS and lasting 8 times longer) we get:
– GLS operation phase: 4.86 mg – 66% = 1.65 mg (as long as EU permits unfiltered coal emissions) = total 1.65 mg Hg on average. (In countries that don’t use fossil fuels for electricity production, like Luxembourg, Iceland, Norway, Sweden & Switzerland, the sum total is 0.)
– CFL operation phase: 1 mg – 66% = 0.34 mg + scrapping phase (assuming no recycling): 4 mg = total 4.34 mg Hg.
In other words, when feeding correct numbers into the calculation, we get the opposite result!
1. Mercury: A Broader Perspective, IAEEL Newsletter 3/93
2. EIA: Electric Power Monthly, September 2009
3. Life Cycle Analysis of Integral Compact Fluorescent Lamps, 1991
4. More on mercury, IAEEL Newsletter 1/94
5. Lot 19: Domestic Lighting Part 1, Chapter 4
6. Eurostat: Panorama of Energy 2007
VITO LCA study
Update 14 Sept: Before the ban, Dutch consultant firm VITO were hired by the Commission to make a very extensive and detailed life cycle assessment attempt, which had the potential of straightening things out. But as far as I can tell, it appears to contain such serious flaws as to make the its final conclusions highly questionable:
A. Using unusual lamp wattages (54W GLS and 13W CFL) for base-cases, both with incorrect lumens for that wattage-class.
B. Putting clear and frosted GLS in separate classes despite the difference in output being virtually non-existent and all other things the same, while the widely varying CFL models (bare, covered, dimmable, outdoor, daylight, improved CRI etc) with their equally varying efficacies, applications and life spans get represented by one (!) class and CFL type only.
C. Incorrect (too short) life span for typical low-voltage halogen lamps, skewing comparison with other lamp types.
D. Overly optimistic estimations of CFL recycling rates (“20%” in all of EU).
E. Like most pro-CFL ‘studies’, this one does not count the mining process for the mercury and phosphors (stating a “lack of info” on that part of the process).
“To produce purified mercury in a CFL, the extraction process releases about 0.4mg for every milligram produced into the waterways, atmosphere, and soil as waste. This is a well-established worldwide average that includes many processes, both crude and hi-tech. This means that the 4mg in the CFL actually represents 5.6mg of mercury that enters our environment.” 
F. Making distribution impact estimates on the assumption that all lamps are produced in Europe, while fully aware that most CFLs are produced in Asia:
VITO: “The distribution phase contributes more than 5 % of the life cycle impacts for 11 of the 15 environmental impact indicators. Impacts of this phase are the highest for the emission of PAHs (69 %), heavy metals (22 %), volatile organic compounds (VOC) (21 %), and particulate matter to air. This can be explained by the assumption related to transport in trucks from the retailer’s central warehouse to the shop. (…) according to the MEEuP methodology (section 5.3.6, page 96), a mix of means of transport (trucking, rail, sear freight and air freight) with assumptions on distances was used for all base-cases. This assumption could be considered as disadvantageous for lamps mainly produced in Europe (e.g. GLS-F and GLS-C) and advantageous for lamps produced in Asia (e.g. CFLi).“  [emphasis added]
G. Not including the energy used to recycle the mercury.
“Collected CFLi’s at end of life are crushed in a closed installation and sieved. The mercury containing fraction is distilated at 600°C to separate the mercury. The pure, metallic mercury is used again by lamp industry.” 
This process seems more complicated than it sounds, and must require a substantial amount of energy too :
H. Not including all the forced individual driving to remote recycling stations for householders who wish to leave their CFLs for recycling, or to the few retailers who have a recycling program, and then from them to the recycling stations, then transportation from recycling stations to reprocessing factories and from reprocessing factories back to the lamp factories. As Peter Thornes points out, when “the lamp industry” has their CFL production located in China, that’s where the mercury has to be shipped back to.
“However, it is not just the energy requiring manufacture (after all, CFLs have longer lifespans, which gives some compensation). It is also the greater emissions from their longer transport from the fewer centra in which CFLs are economical to make (China), and it is also the further CFL transport emissions to recycling plants and the emissions of their reprocessing there, and the further transport of reprocessed parts to different locations.
This means that inter-continental transport between China and North America/Europe can take place twice, since CFL content including mercury may be shipped back to China for reprocessing and new manufacture. Even more significantly, shipping use of bunker oil, the worst CO2 emitting type of oil, greatly increases the emissions involved (more)
Sounds like an awful lot of driving, shipping, processing and polluting, doesn’t it?
1. Mercury Risk in CFLs: The Facts
2. Lot 19: Domestic lighting Part 1, Chapter 5 (pdf)
3. Lot 19: Domestic lighting Part 1, Chapter 4 (pdf)
4. Technical guidelines on the environmentally sound management of mercury wastes (pdf)
5. New Electric Politics: Life Cycle
Osram LCA study
Update 12 dec: See also my post on the OSRAM Life Cycle Assessment study.
Update Aug 2012:
Here is a ‘How it’s made’ videos the simple manufacturing process of incandescent lamps, easily done in a local factory:
And a commercial video from a CFL factory in China, showing the infinitely more complex and potentially hazardous manufacturing process, complete with picture of an oil tanker by which finished bulbs are shipped from China, no doub gulping oil and spewing out quite a bit of pollution on its way to the West.