That last step of the logic is not correct. The default benchmark is whatever calculation has to be completed before getting a score, but obviously in this little world we try to optimize in any way possible to get a better score. We
- set cores manually in Wprime
- change run order in 3DMark01
- adjust affinity settings for Aquamark3
- disable feature tests in 3DMark Vantage
- ...
The UCBench 'benchmark' does not change when only selecting SSE3 or manually adjusting thread count, hence both are legit. It does change when you would change the bench-mode to time-based or password-based, hence it's not allowed to change that.