This really brings into focus the question of whether any of the "varifill" code should
really remain. This was probably one of the only remaining places where a partial-fill
operation was done and the code was broken for this case. What we know about disk i/o
streaming doesn't support varifill much at all. Something to think about.