could you try running sciencemark2 clock for clock? it would be a nice benchmark to represent crunching and you wont have to deal with different WU's.