there seems to be points of diminishing returns for each of the clocks (core/shader/mem) when maxing them all out together (i think)... so it might be wiser to take data incrementing each of the clks individually using 3dmark06 to see what portion of the score increases.. then see if you can find a happy medium for all three.

example, i know i can clock my core over 770 and my shaders over 1944, and my mem up to 1064 with stock cooler, but using all these together might result in a lower score then using lower clocks. I'm currently taking data for each clk at each increment, individually, to see if i can find the best combo for my card.