Thanks Michael,

The file contains the rotated curve and a copy which has been distorted slightly. It seems that as soon as the points are moved at all, even if the curve is not tight, shelling/offset won't happen.