Here's a long shot...(may not make sense at all!!)
First you should run 3dOverlap to actually determine how many voxels overlap between the 2 studies.
Then you "could" use a chi-square test.
You add the number of voxels in each cluster of each study (eg, 25 + 30), and then, say the overlap is 20 voxels, you can compare the likelihood of the Observed overlap (20) vs the Expected Overlap (55).
...my two cents