I think benchmarks are shit too, and they favor Intel Macs over PPC Macs somehow.
Everything is optimized for x86, many times more effort has been put into optimizing x86 compilers compared to PPC compilers. Imagine if the same effort was put into PPC compilers.
Legend says that one day a hero, True of Heart, shall seek and find the One True PowerPC Compiler and pull the installation DVD from the stone block in which it's embedded. Once this noble deed is achieved a New Age Of PPC Wonder shall dawn, and the world shall finally be cleansed of all sinful x86 wickedness. Rainbows will dance all day with the sun in the sky, and Unicorns and Care Bears will once again frolic with the My Little Ponies in our pastures. The rivers shall overflow with fish, the forests with game, and the fields with wheat and plenty. And Steve Jobs himself shall descend from the heavens and walk amongst mortals, enlightening them all with love, understanding, and shrewd marketing dogma forever and ever and ever. Amen.
It's amazing how thoroughly Apple managed to brainwash its customers with the "PowerPC is better" gospel. I'm sorry, but it just isn't true. There just isn't this huge conspiracy out there constantly working to make PowerPC CPUs suck at application benchmarks. (If anything, the infamous Photoshop benchmark suite that Apple used to advertise as showing that G4s were *so much faster* then Intel's Pentiums was heavily rigged.)
My take on the performance of the G5 from the time I had with one is that the CPU itself is a competent number cruncher. The 2.0Ghz Xserve was significantly faster per clock then the (P4-based) Xeon servers I had to compare it with on various UNIX-y floating point benchmarks. (The 2.0 G5 could edge out a 2.53 Ghz Xeon on most tests and came close to the 2.8s.) Also just for laffs I sic-ed it at SETI@home as well, and I recall it being able to churn out a work unit in a little under 3 hours on average while the Intel servers were more like four. (the exact times I'm a little fuzzy on, it's been a while.) These were all tests that measured the output of a single CPU. So when G5 is compared to P4/Netburst it comes off pretty well. However, the first time I tested an AMD Opteron it was close to a dead heat on the UNIX benchmarks and at SETI the Opteron could shave almost another half hour off the work unit time. The Opterons ran at basically the same clock, so... the G5 as a number cruncher is good by 2003 standards. Not "the best", but still pretty awesome. It's not 2003 anymore and the Intel Core series is better per clock then the Opteron, so the G5 is looking sort of mediocre now. But it *was* good.
The real problem with the G5 is it's a lousy desktop CPU. It's floating point performance was good/awesome, but its integer unit is comparatively weak even compared to the G4e's, let alone the P4's. (Look for the Ars Technica article about the PPC 970 for more details. Its integer unit was basically lifted straight out of the Power4, which was optimized for a huge frontside cache and properly-ordered 64 bit code. The 970 had a much smaller cache, and of course Mac software was distributed as 32 bit binaries. Recompiling can help with the latter but not the former.) Integer performance is what counts for desktop performance, and the G5 didn't have it. It also of course had poor thermal management compared to CPUs designed for desktop applications. In the Xserve it seems to work fine, but then the Xserve sounds like a bit like jet engine even sitting idle. Apple's constant problem with the desktops was balancing fan speed against CPU throttling, and they just never figured out how to make the machines consistently quiet enough without them either crashing or performing terribly.
Finally, it was a big mistake for Apple not to include ECC memory support in at least the "Pro" desktops. IBM has good engineers, but they come from a background that assumes that anything with a large amount of RAM uses ECC as an engineering best practice. Apple felt they "needed" to save the 15% or so ECC adds to the memory price tag so they special ordered a Northbridge not supporting it for the desktop machines. (Smart move, guys.) Intel *should* probably push PC makers into using ECC in their desktop machines, but since they don't they've gotten pretty good at making memory controllers reliable enough to do without it.
In short, the G5 is "exotic" technology crammed into an environment where it doesn't really fit. It's akin to, say,
the Chrysler Turbine Car. It's certainly possible to scale down a jet engine to replace a reciprocating internal combustion one, but it takes a lot of engineering to make it work, and the end result might at best be only *arguably* superior to what it's competing with. ("Is it more powerful? No. Is it more fuel-efficient? No. Is it more reliable? Probably... once we get the bugs worked out.") If Apple's/PowerPC's desktop market share were such that it could sell as many CPUs as Nintendo and Microsoft can sell game consoles then PowerPC would be alive and well in the desktop market. Lacking that, IBM/Freescale don't have anywhere near the patience or goodwill necessary to spend their own money designing something very expensive for one noisy customer who *might* be able to sell a million units a year, and Apple couldn't/wouldn't pay enough to make them do it. Instead, Apple went to Intel so they could stop throwing money down the sewer every time they needed a new generation of CPU performance to keep up with the competition.
Which was the right choice. Instead of worrying about hard technical problems Apple can keep doing what it does best: chiseling blocks of aluminum into fancy shapes and marketing the living **** out of them. :^b
Anyway.
The system's problem is really weird, for instance, I've had the machine up and running for days, almost a week even, and it's been fine... then restarting the machine booting into leopard will make it crash! xx( ... taking out the user upgraded (Kingston crap) seems to make the G5 systems more stable but slower...but it'll still misfire on occasions plus there's no LED warning light inside (unless i'm looking at the wrong place) but i'm not really sure right now. Would like some diagnostic software to prove it's... what's available?
It's sort of a dumb thought, but have you tried removing the stock ram and running it just with the Kingston crap? An article I saw about G5 desktop reliability said that some people had experienced the stock ram getting flaky over time. I'd be skeptical, but it's worth remembering that PC3200 RAM was quite new on the market when the first-gen G5's came out. A lot of what was sold as "PC3200" was essentially factory-overclocked.
If it's consistently always crashing shortly after booting that's an "interesting" symptom. Are the crashes almost always after a cold boot? If it tends to be reliable once warmed up that might point to something not being seated well and thermal expansion being an issue. It's also possible that some piece of motherboard hardware is getting flaky and failing initialization sometimes. Do you use wireless? I've seen bad Airport cards cause crashes after bootup.