Neo, the way it was explained to me, is that unlike F@H, which uses the flags to tell it use certain features of the cpu (SSE2, etc) to actually crunch with, in Boinc programs other than Seti all it does is use it for benchmarking and not to actually do any work with.
I'm finding out that whether they use the quorum system and how many they use for a quorum, will even out some and make others very tilted.
Here's a question to ponder- Linux has a bug that gives an artificialy low benchmark, so they recommend you "optimize" Boinc for Linux. That brings up to an even level with what the same machine would bench if it ran Windows. But even with which version of Windows Boinc? The official Boinc version, I'm assuming. Meaning that those running and optimized Boinc in Linux think they are running on a level playing field, but if they are up against optimized Windows Boinc, then they are in fact being benched at about 1/3rd what the optimized Windows Boinc people are benching?
edited for typo