I'm inclined to believe Google on this one, having seen more than my share of conflicting measurement studies. The CF detection companies did mistakenly classify back clicks on a browser to a landing page as click fraud, among other things (assuming these clicks weren't faked). However, this doesn't mean the amount of actual click fraud isn't at least as much as the CF detection companies are reporting. In addition, Google offered as defense "our logs say differently" (without actually offering their logs as evidence, not that we should necessarily believe them).
I was checking ClickForensics, one of the CF detection companies, and noticed that they have commissioned the 'tute to do an econometric study of the likelihood of click fraud occurring based on ad spending. While this is certainly important, I think there needs to be an equal focus on the actual weaknesses in CPC (and other per-web activity) models that enable click fraud to occur. As I mentioned before, not enough attention is paid to analogous models, such as phone fraud.
In any event, I filled out their contact form asking for more information. The last time I asked ClickForensics for information, I never heard anything from them. This time, I indicated in the comments that I am an MIT graduate, in addition to describing my AltaVista work. Hopefully they'll respond. I'm disappointed at how closed this community has been so far (for arguably good reasons, lest fraudsters figure out what they're doing). I wish it could be more like the IETF which is open to anyone.