Hi Wolfram,
After taking a look at your data, the problem is subtle, unfortunately.
The effect of such a large blur is that the range of values becomes
much smaller at any given voxel (i.e. the min and max of each time
series is much closer together).
That is not itself a big deal. But things go bad because the data is
stored as shorts. So everything is truncated to integers, and the
time series will become a little blocky (the curves are not so jagged
anymore).
When converting the data to floats before the 12 mm blur, the resulting
FWHM of the -detrend estimates as about 13.3 mm. When not using
the -detrend option it was 18-22 mm. Ick.
I'll have to ponder this a bit more...
- rick