Because there is a bug in windows 8 and newer windows OS's where if you were to adjust the BCLK of the CPU, the timing goes all out of whack and the results gained aren't the actual results.
Hence my previous point.
But to answer your question I will absolutely be using legacy OS's if they offer better performance in benchmarks, that's exactly why people still use XP today for benching.
If you had a similar system to XTU where you got a score at the end of running prime 95 that would eliminate those who only up the results at the end as they wouldn't gain anything from doing it. That's if the score isn't calculated in the exact way as XTU of course
Or have a program that just gets crunching number and gives you the total amount of numbers crunched after the 24 hours is up? Kinda like a time based 32M?
Just thinking out loud really.