InfoWorld is running a series of articles on the "Six great myths of IT".
Myth #5, "Most software projects fail", hedges by saying that it "depends on how you define failure", but then provides hard data by citing a Standish Group survey that says that only 34% of software projects are "unqualified successes", 51% are "challenged projects" with "cost overruns, time overruns, and projects not delivered with the right functionality to support the business, and the remainder are unqualified disasters. Sounds pretty dismal.
The graph accompanying the article is particularly interesting. Very large projects appear to be nearly guaranteed to fail. Shouldn't be surprising to anyone in the industry. It's hard to overcome the complexity and lack of accountability in very large projects.