The smoothing procedure is described in the Hutton et al paper "Voxel-based cortical thickness measurements in MRI":
The spatially normalised VBCT maps were smoothed using a
three-dimensional Gaussian smoothing kernel with FWHM = 3 mm. A
smoothing kernel of this size was chosen to reflect the average
thickness of the cortex and to account for any small discrepancies in
spatial normalisation. Gaussian smoothing slightly reduces the values in
the VBCT maps because it performs a weighted averaging over all voxels
included by the Gaussian kernel and the values of some of those voxels
are zero if they do not consist of grey matter. This effect was
corrected for by dividing the smoothed VBCT maps by a binary mask of
each VBCT map which has had the same smoothing applied.
Essentially, it involves tissue-weighted smoothing as described in this presentation (except the presentation describes it for WM instead of GM):
http://www.fil.ion.ucl.ac.uk/~john/misc/VBQ.pptx