Thanks for finding (and fixing!) this bug, which I inadvertently introduced while changing the square sampling to disk sampling. I've checked the change into CVS, which should show up in the HEAD in a day or two.
From: Mark Stock <firstname.lastname@example.org>
Date: May 14, 2010 8:25:23 PM PDT
I was playing with the -pd option in rpict in Apr 27th HEAD and noticed that the blurred portion of the image was always skewed to the up and left. See the following movie: