Thank you, Rick, for those suggestions!
I'd like to ask a little bit more specifically:
A) How can it be that the estimated FWHM is less than the applied FWHM?
(( To exclude the preprocessing from the question: I now took a functional dataset and just blurred it with 12mm FWHM. That increased the FWHM estimated by 3dFWHMx from 4.5mm to 9.8mm (using a whole brain mask for 3dFWHMx; without a mask it was increased from 4.3mm to 7.4mm). I am surprised that it is not increased to at least 12mm. ))
B) Which FWHM should one give to 3dClustSim when the estimated FWHM is less than the applied?
(( I'd tend to give the maximum of the two numbers. ))
Thanks,
Wolfram