Regarding what Bob said using games - this would be nice, but I only have one spare computer for testing, and don't have the patience to wait three or four days for results every time I make a change.
But, Dann Corbit has posted a really good set of "positional" tests recently, and I'll use that.
I was thinking of using total time to a fixed depth for each position (e.g depth 11 on all positions). Using a batch file I can get a comparison of several changes in a few hours.