Right, I've been struggling a bit more with this thing. When I said that softedges weren't doing it for me I basically meant that I thought they were a hazzle to work with.
After beating myself over the head all night, I realized that it was easier to first do the rotoshape inside the motionblur all through the sequence and then come back and pull out the softedges.
Now this is easier and is giving me results that are approaching something acceptable, but I'm not really happy with it.
So a more appropriate question might be: What is your roto workflow when you have a lot of motionblur in your sequence?
Feature request:
A way of distinguishing between hardegde point from softedge point in a rotoshape.