It feels like people on both sides of this "argument" (though it should be a friendly debate, it doesn't feel that way) are sort of missing the point.
As far as the way these particular benchmarks are executed, there's nothing inherently wrong with the tests. They took three commonly used high-end programs and a game, ran them on factory standard high-end boxes, and published the results. The tests were basically fair for what they're testing.
That said, the tests fail at two things:
1) If you're wondering what the theoretical capabilities of the chips in the computers are, these tests are useless.
Since none of the programs are particularly well optimized for any of the really new processors, they don't tell you what either the AMD or IBM chips are really capable of when allowed to run all-out. And, only two of the four programs (Photoshop and Quake) use a codebase that is well optimized for the Mac at all; Premiere isn't well optimized for the Mac (it's now a dead product, in fact), and Word uses a completely different codebase on the two platforms.
Most of Apple's benchmarks were on processor-intensive number-crunching programs that were optimized at least for the G5 (like the NASA benchmarks published a while ago). Under those circumstances, I'd expect the G5 to be at the very least quite competitive with AMD and Intel chips, as this article shows it in the optimized-for-both-platforms Photoshop.
2) If you want to know how productive you will be with an Athlon or G5 based box, these tests are nearly useless.
If you use any of the four applications they tested, then these results will apply to you.
HOWEVER, most people doing video editing on the Mac don't use Premiere, because Final Cut Pro, by most accounts, demolishes it--I know which one I'd choose. They can't test that side-by-side since FCP doesn't even run on the PC (which might be reason enough to opt for the Mac), but a better speed comparison would be to perform a similar operation on Premiere on the PC and FCP on the Mac. Premiere might still win, but at least then you'd be making a comparison that users would be more likely to benefit from.
Photoshop is heavily optimized all around and a widely used app, so that's a much more fair comparison. The G5 did well there, I might add.
I'd expect Word to be vastly faster on the PC (it is MS, after all), and it is. [Of course, it's pretty darned fast on any platform, so I don't much care--not many of us do find-and-replace on a 1500 page document, and even then you can probably wait the extra 15 seconds for that, about five of which look like they came from the lack of a RAID array.]
If you're doing hardcore gaming, you'd be an idiot to buy a Mac for it, but the G5 actually did quite well in the Quake III test--it held its own, particularly against the other 128MB cards and the P4 system.
What I'm getting at with this whole rant is not that the benchmarks are unfair, or intentionally fudged, or meaningless. But it is a fairly narrow spectrum of tests, and in my mind only the Photoshop and Quake tests really show me anything I want to know about the G5 versus Athlon 64.
All things considered, I think the G5 looks pretty good.
