Linus makes a good point.

I think this should work:

1. Take two comparable systems (external bus speeds and external clock are the same). Measure base SPECint2006 of both.

2. downclock the system with better score until SPECint2006 scores are equal.

3. Compare power use.