At first we have to determine which card will be the chosen one to be tortured by us. And for benching, in my opinion we should use the benchmark that is most dependent on GPU-power ( that is 2k1, isn't it? ), and maybe limit the CPU-frequency to make it at least a bit comparable ( most users will use a Core2Duo I guess ).

Some thoughts:
1. CPU frequency fixed to ???? ( 3GHz?! )
2. dual cores and quads? different categories?
3. categories determinded by core temps of GPU ( as suggested by STEvil )

What do you guys think?