You have to give this guy some credit for at least trying, but the standing problem with these types of studies is that there are just too many variables. For instance, this particular study only consisted of one program. Also, it doesn't really control for program proficiency very well...he tried, but environmental impacts are going to skew those results (classes taken, work environment and language of choice, sample set bias for the previous, etc). Also, as a previous poster mentioned, these are from 2000. Since 2000, Java has made serious performance ground. I would be interested to see a few more studies like this...even if they aren't perfect, they do tend to remind people that there is a huge difference for language productivity, and that it dwarfs performance.

Maybe we could get a version where the program was to implement a decent sized project requiring several developers, to emphasise the code quality angle? That would take a lot of bored programmers to do, though.