run 3dmark with the best cpu you have as it will be least cpu limited, however bear in mind that you might still be limited in 3dmark with stock clocks, its impossible to say really, though you could if you have the pro version change the res/aa/af levels so its completely gpu limited, if you're hitting 12k now and you up the res so the 8800gtx is only hitting 6k you know you aren't cpu limited type situation, then compare that to the r600 at same resolutions.
there are games you can test, it really doesn't matter all that much, just up the resolution and quality levels until the fps drops enough to know you're mainly gpu limited. its hard to say what levels of performance to expect. hl2 stress test or the cs:s test(can't remember which the test is in) should be fine, you can use a 8800gtx to get 250fps at 1680x1050 with every setting max but little to no aa, and drop that down hugely with aa enabled to very high levels. testing at that no aa setting is pointless as its mostly cpu limited, but with 16xaa you're looking at 130fps maybe, and you know the system can cope with 250fps, so you're not cpu limited.
how drivers cope with games, how new drivers which, lets be fair, if vista is extremely hard to write dx10 working drivers for, and nvidia has a 6 month head start, its not completely fair to look at first set of drivers and gage performance, IQ and buggyness. they might take 6 months to fix like nvidia, they might have it all fixed in a release a month later, though at least you DO know there WILL be a release a month later.
testing a couple games on XP aswell as vista, vista to see if drivers are good, XP really should be a "little" easier to show a hopefully more fair comparison. also, i think nvidia's dx10 demo has a fps counter in it, there are a bunch of options in the thing, if you can record a demo to replay on diff card and compare fps, or if you try and just do the exact same things. its some kind of waterfall test thing, you can add water sources, if you do test it the best way would be start same place, zoom out as far as poss and check framerate, zoom in on same point/angle, check framerate then add as many water sources as it allows and test framerate, do same on both cards and compare. though obviously nvidia may have set a trigger that it can only run with nvidia drivers present or something of that nature.
only 1/2 games you'll find quad core better than dual core, 3dmark, sup com, a couple flight sims and i think thats it, everything else, same cards am2 6000 vs quad core you won't see a big difference at all.