Superior performance which, thus far, remains officially undocumented. Where's the SPEC results, if it's so hot? Even Apple appears to have pulled those fakey-SPEC rate results from their website now.
From my own look at it, it appears to me that a single G5 is approximately on a par with a single P4, on a clock-for-clock basis, and so far I've seen nothing to disabuse me of that notion.
Anyway, for distributed computing - particularly distributed computing on someone else's computers, where I don't have to worry about how much they cost - individual node speed really doesn't matter, only aggregate performance, like I said. Given that there are five times as many HardOCP members as there are OSX members, the average PPC box will have to be five times faster than the average x86 box for Team OSX to match the performance of Team HardOCP. And that's just not happening. Even if the PPC team triples the work-unit-per-node performance of x86, which is wildly unrealistic, the fact that there are so many more x86 boxes still puts them way ahead. Team OSX may not fall behind as rapidly with new compilers and the like, but I expect that they'll still be falling behind, if for no other reason than because they're simply outnumbered.
Now, if we're shopping for a supercomputer, that might be a different story, but that's not the example we're presented with here.
I'm waiting for SPEC too. Looking at third-party PPC/Intel results shows a dual G5 2 GHz roughly equal to a dual Xeon 3 GHz, less in integer and better in FP. No way is a PPC 970 1 GHz as slow as a P4/Xeon 1 GHz. The processor architecture is just much more advanced, among other things capable of keeping 215 instructions in flight at a time.
The Xeon/P4 has a big problem with bus, as shown recently in an Opteron/Xeon shootout, where the Intel's slow, shared bus can't keep the processors fed. Plug 20% faster Xeons in the current system, and you will barely get more speed. Intel is slapping lots of L3 cache on the chips to alleviate the problem, but the end cost of the chips is enormous. The G5's bus runs at half the speed of the processor, and each processor gets its own bus, so the G5 will scale well to 3 GHz this fall, with a 1. GHz bus to keep those processors fed. Intel seems to be hitting a wall.
Given that there are five times as many HardOCP members as there are OSX members, the average PPC box will have to be five times faster than the average x86 box for Team OSX to match the performance of Team HardOCP.
That's why distributed computing projects make mutltiple versions.