Hi Anderson,
I update PALM, (originally it was alpha56, now it's alpha79) and run the design with 200 permutations. Unfortunately the problem remained. Yes everything looks good in the FWE results. But if you look at the "vox_zstat_uncp_c3", it is a binary map with the value of 0.995 in vast grey matter areas. Actually, this was the map I was talking about in the original question email. Sorry I didn't make it clear.
Thank you
Kai