Thanks for your advice so far Rich and Sally.
A couple details: I am dealing with a correlation map. I have tried to raise the p-thresh, but at 0.00000001 I still get the same cluster. I believe (and according to other studies from this lab) the valence of correlation across this "cluster" is not uniform (cluster begins in the inferior frontal gyrus, goes through the subcallosal gyrus, then through the ventral striatum, and ends somewhere in the parahippocampal gyrus).
Ultimately, I think that I will probably use the draw dataset plugin. But I really wanted a justification for *how* I broke up the cluster. There are few places where the cluster becomes incontiguous in one dimension, but in another dimension is still part of the megacluster. The idea of finding the center of activation is intriguing and seems justifiable, but how...