Hi Michael,
I would suggest making a copy of your blurred data for a single run,
applying the additional blur to it, and then running 3dFWHMx to the
result (if you have the errts time series, applying there would be a
little better).
Then average those results across subjects.
That should be sufficient. If you are applying a gaussian blur to
the results, the effect to the FWHM values should be basically the
same as if you had applied that to the data ahead of time. Only
measuring the blur on the results is not accurate (except for the
residuals).
- rick