That FFT is probably written in VBA for Excel as a plug-in, yea? That's likely to be at issue for starters, and if the 1024 points are coming out of the spreadsheet and each point is being passed to the VBA individually (rather than as one vector all at once), yea, I can see that taking quite a little bit of time.

On any modern PC, I think that if I were coding FFT's the way we used to (in lovingly crafted assembly language for speed), my gut tells me that a 1024-point FFT should be able to be done in under 100usec in single precision, under 250 in double precision. (Edit: brain fart -- forgot that 'm' was milli, 'u' is micro)

And those are conservative estimates.

The gamers have quite the tool at their disposal:

http://www.cs.unc.edu/~geom/GPUFFTW/results.html