rharder, have you tried using "register" keywords. I played around with it, but I don't think that I was doing it right because it did not seem to increase the speed any.
One funny thing is that I went and changed all of the doubles to floats just to see how the performance changed. It was actually slower!!!
Side note: Apple is rumored to be engineering the G5 with the roadmap saying it will be up to 2+ ghz.