John Thacker points us to a study of California (registration required, but free). I've only browsed it but the introduction states:
In September 2008, the California state legislature passed the first state law (Senate Bill 375) to include land use policies directed at curbing urban sprawl and reducing automobile travel as part of the state’s ambitious strategy to reduce greenhouse gas (GHG) emissions. The legislature recognized that cleaner fuels and more fuel-efficient vehicles would not be sufficient to achieve the state’s goal of reducing GHG emissions to 1990 levels by 2020. The bill requires the state’s 18 metropolitan planning organizations to include the GHG emissions targets established by the state Air Resources Board (ARB) in regional transportation plans, and to offer incentives for local governments and developers to create more compact developments and provide transit and other opportunities for alternatives to automobile travel to help meet these targets. ARB currently estimates that reductions in vehicle miles traveled (VMT) resulting from these actions will contribute only about 3 percent of the 2020 targets–an estimate that reflects uncertainties in the state of knowledge about the impacts of more compact development patterns on travel and the short time horizon involved.
In other words, the environmental benefits of checking pro-suburb subsidies are real, but they are smaller than many people think. That's from the National Academy of Sciences and the authors are no haters of the environment. If you check out p.59, you'll see that a forty percent increase in population density decreases vehicle miles traveled by less than five percent. pp.131-132 offer a summary of the study's conclusions and quantitative estimates.
The authors conclude that density-friendly policies are a good idea, and I agree, but still these are not overwhelming effects. Keep in mind that current trends are strongly pointing toward population dispersion, so to reverse those trends and see greater density would take some doing. We're not close to that margin.
There are many other interesting parts of the report, including case studies of Portland and Arlington.