These scores look good. Keep in mind that PSBench calculates the overall score in a mindless way (summing the individual times) that puts a vast majority of the "total score" weight on just four tests: accented edges, pointillize, water color polar, and radial blur. Coincidentally (or maybe not...conspiracy theory?), these are the 4 tests in which the G5 does particularly poorly, so the fact that it beats the Athlon 2200+ in the overall score is actually quite impressive. Basically, you should think of the overall score as just consisting of those 4 benchmarks I mentioned.
A much better "overall" benchmark is to look at the number of tests each machine won (though this is still imperfect - in fact, there is no perfect benchmark in this case unless you know with what frequency you do each type of operation). Ignoring the lone DP machine, the P4/3.06 HT won the most tests (10) - no surprise there. The G5/1.6 won the second most (5), and the Athlon 3000+ won the fewest (4). I would not be surprised if overall the G5/1.6 could match the Athlon 3000+, though still fall somewhat short of the P4/3.06 HT (but then you are comparing bottom of the line Mac vs. top of the line PC, so what do you expect?).
