Now that we can use concave geometry in WED, I had thought about doing it this way some time ago. I'd just build a super complicated/detailed level complete with shaders or whatever else I need and take the appropriate screenshots. I'd take care of the collision hull by reducing the geometry and leaving in the more detailed sections for the shadows. If I want to get a render of specific sections for my layers then I just turn off the parts I don't want showing and do this for each individual layer.

After that it would be a process of using panels. Everything should line up nicely. However, using panels would cut out the ability to move a realtime camera into or out of a level. I suppose you could fake it by scaling panels in realtime to make it look like you're getting closer or farther away from certain sections, but that would probably look terrible. That's why I'd want the layers in the actual level itself and not use panels for background/foreground art. This is where the pain comes in because you have to manually line everything up and hope it's to scale. The background plate would be easy but the foreground elements or things you can walk in front of or behind are the tricky ones.