Thanks for the link-
The deep-maps are an interesting idea, but generating the 1d functions that layer into the shadow map seem not applicable to today’s real-time hw methods. Currently we’d need to use software rendering to generate the 1d functions that are indexed by the 2d shadow map. I think if we could render to 2 targets simultanously(something DX10 is rumored to support) or if it were possible to render to a 3d texture (not rumored/supposed anywhere), then we could approach the amazing benefits of occulsion percentage/ transparency / volumetric shadowing.
Overall though, i think that the work done in Depth Peeling is more applicable to current real-time handling of transparency in shadows (multiple depth-buffers).
As always, Pixar leads the way into some really quality pre-rendered visuals (how about that hairball?!?!)