ah wait yes, I missed the part of it being a duplicate mesh with no preparation needed.
so the ellipsoid is applied to hide the chopped part on the main guy, and then for the duplicate mesh everything is hidden except the ellipsoid.
so now I see the beauty of your system: you can chop body parts in any amount of locations and the only preparation needed is an additional PhysicsAsset to match those locations.
I’d even argue this step isn’t needed! in your proxy mesh you could use the full PhysicsAsset, through code find the phat bodies that are not relevant to the proxy and disable collision for them. all you need is to keep track of the HitBone when the dismemberment happens 
but as you still need cap meshes to hide the holes, you still need pre-determined chopping points that use pre-determined cap meshes. unless there’s a way to adapt the cap meshes to the holes, wasn’t some mention of this in the L4D paper?
I’ll have to re-read the concept behind your ellipsoid system but the simplified workflow starts to make it more appealing to me compared to the BoneBreak system.
random thought: what if instead of ellipsoids you would use a HitMask to determine what gets chopped?
btw in the UDK system you’d also be able to make a hole into the torso and then decapitating the same character. you would need your mesh to be prepared in such a way, pre-making every possible hole (and not allowing overlapping holes). but yes this is too much preparation work 