My comparison will always be games. They stress the computers more than anything else.
Given G4 or G5 or whatever vs the top PC, with an equal video card, the PC has always done better.
I'm going to sound like an apologist, but there are good technical reasons for why the above has been the case. In some instances it has been as simple as the port to the Mac being done poorly. There are some obvious reasons for this, such as not converting the MMX/SSE optimized x86 code to Altivec on the Mac, and using the slower non-vectorized paths, and other not as obvious ones such as a larger penalty on int<->fp conversions on PPC (so a lot of x86 code unnecessarily does these conversions because it's not a hit on their HW, but slows down when you run the same thing on PPC). Another reason is many games are originally written for DX, and then run through a translation layer to run under GL on the Mac since there is no DX on the Mac. Anyways, the bottom line is that due to a multitude of reasons the games that are not natively written on the Mac often suffer from "PC-isms" that slow it down. Many of these issues could be addressed if the developer doing the port spent more time optimizing specifically for the platform.
Aside from porting issues, there are also issues with the graphics pipeline and what types of shortcuts and tricks the drivers take. Ever wonder why the "pro" drivers for ATI and nVidia (Quadro/FireGL) run games slower than the "consumer" drivers? At the same time, why do the "consumer" drivers run the pro apps slower than the "pro" drivers? Apple has taken the path of pleasing both sides of the camp resulting in slightly slower, but more comprehensive, more accurate rendering in the GL layer.
Anyhow, we've done extensive testing in this regard for every game that runs slower on the Mac to isolate and understand exactly why it's slower. For games that are properly optimized at the same level as the PC version, the difference in performance is usually negligible or nonexistant.