Hello Cristina,

There is no magical solution as you realize. There is quite a bit of variability in the volume or extent of activation in BOLD. Forgive me for tooting my horn, but you might find this paper of interest: "The Spatial Extent of the BOLD Response" Neuroimage 19, 2003 pp 132-144.
But even if you have a large variance, you can still tell the difference in activation volume between the two groups however that might require a large number of subjects.

Speculations:
1- Another approach that I suggest you try is to focus on a few regions of interest and ignore the rest of the brain. Perhaps by doing so, you can lower your threshold far enough (because of the smaller number of multiple comparisons) where the patterns of activation are less variable. But that all depends on the noise levels you are dealing with.

2- Also, you might want to try looking at changes in the spatial patterns (rather than volumes) between the two groups. You'll still need to restrict this analysis to small ROIs. See Jim Haxby's paper: "Distributed and Overlapping Representations of Faces and Objects in Ventral Temporal Cortex", Science, Vol 293, Sept 01.


cheers,
-ziad