Well what im describing is a bit different i think? It would be a real shadow buffer.
You would draw the scene twice clip alpha drawn pixels to get a a rendered z depth in two render targets.
These would be depth samplers for the scene when drawn normally with a pixel shader.
(the cavet is that the normally drawn scene would calculate polar coordinates to get the sampled depth then comparing that to its depth from camera so it would be exactly how you shadow in 3d but in 2d.
The depth would conform to the shape of the sprites themselves Not the corners of polygons that contain them or any generated polygons. Basically in 2d you would be finding edges of whats in the sprites not the sprites polygon.
edit this is the idea.