Performance isn't everything. For example I'm wondering why Michael haven't included any Unigine benchmark... because guess what, r300g kicks fglrx ass there (mainly because fglrx can't run those and r300g can just fine). Also r300g is advertising way more extensions, including some from GL3.0 or later. And don't forget about stability. Back in the days of fglrx I got xserver crash at least once a week.
On the other side, if you spot some performance problems, open a bug, they _are_ being fixed! From almost 60 bugs I've opened at fdo in the past few years, only 9 remains open, sometimes being fixed only a few hours after being reported.
For example there is at least one, maybe two performance regressions judging from reading Michael's results, and some misrenderings, which he is probably never going to report, and the developers won't never know about this and never fix this ( well actually I do realize that there are some developers reading this forums but I don't expect them to be in a mood to fix anything after reading all those cheerful and supportive comments ).
I'll see if I can reproduce any of this with my RV530 card, but this bugs may as well be specific to r400 cards. This is the reason why getting a wide testing coverage is really important, there are many different card models for every generation and the developers doesn't have time to test everything (I'm not saying they do not test, but piglit runs on one or two different cards is never going to catch all bugs).