Not to sound snooty or anything, but exactly: you don't see the huge difficulty in it. What I'm telling you is that the huge difficulty does exist. In fact, there are a number of organizations devoted to exactly the sort of generalized performance measurement you're talking about. SPEC is one of them. But there are problems with that sort of approach too, as I said. One problem is that Intel has a compiler group which, essentially, specializes in getting good SPEC scores on Intel chips. Nobody uses their compiler for building anything people actually use, but you can't beat the results it achieves on SPEC benchmarks.
And if you want to use SETI@Home for benchmarking, then you're not doing a generalized benchmark at all. The results don't really imply much besides how well the SETI@Home client happens to run on one particular system (not even one particular brand, type or model of system, but one one specific computer).
