When doing this kind of tests, I believe "timed kernel compilation" should be set apart from other benchmarks.
You see, while less is better is what we're after, new compiler versions tend to run more optimizations for the same code. So even if the time goes up, if the resulting executable is faster, it's still a win. The current layout of the article does not highlight this particularity at all.
Jut thought I'd throw my two cents in here.