Hi Rainer,
how about aggregating to a lower resolution, where the cell center -
cell center distances correspond to your minDist? Then sample as one
would normally and then take the cell center coordinates from your
sample back to your original raster.
A bit of geometry will tell you what the resolution needs to be and
please tell me you are not sampling a longlat projection!
Either the r.region GRASS function or the aggregate 'raster' function
should do.
there's my cents/sense...
HTH,
Brian