It's just amazing that this kind of nonsense persists. There were no significant benchmarks, "scientific" or otherwise, at the time or since showing that kind of behavior. The G4 was a dud. Apple rushed out some apples/oranges comparisons at launch (the one you link appears to be the bit where they compared a SIMD-optimized tool on PPC to generic compiled C on x86, though I'm too lazy to try to dig out the specifics from stale links), and the reality distortion field did the rest.