Hi David,
Thanks a lot for bringing up the point about a principled approach for determining the combination of p-threshold and extent threshold for achieving a desired alpha value (familywise error rate). Multiple combinations of p-threshold and extent threshold give the same alpha value. Conjunction analyses (where you try to find overlap of activity across conditions in order to determine multimodal brain regions) can give totally different results depending on the exact combination of p-threshold and extent threshold used for controlling familywise error. But other than conjunction analyses, I have not come across any other analysis where my results are too sensitive to this combination.

I agree with Rick that there are no wrong or right combinations, there are only reasonable and unreasonable combinations, depending on what effect you are trying to tease out. For a conjunction analysis, for example, it will be unreasonable to use p=0.05 and extent=80 voxels even though this might achieve alpha=0.05 because such a combination will almost always yield overlap between conditions.

I also agree that the only real solution to the problem is transparency on part of the authors and a demand for transparency on part of the reviewers.
Best,
Gaurav