Here is some number I can provide. I ran so called Fast Multipole Algorithm which can accelerate the computation time for matrix time vector. I used only 10 source points to save time (it usually goes up to 100,000). I used gcc with -O3 option for both computer
PowerBook G4 1.5Ghz, 256+512MB, 15", 49sec
PowerMac G5 Dual 2.0Ghz, 512MB, 17sec
Also I used -fast option which available only with G5
PowerMac G5 Dual 2.0Ghz, 512MB -fast option 14sec
And Just for fun I tried it on my dell pentium 2.5Ghz with visual c++ complier without any option it took 61sec
I hope it helps