I think I'm done with all the theory. It's not as easy as simple bruteforce z-buffer drawings of course. Was kinda tricky to figure out, but again it proved that my the first actually interesting thought in my mind was the best one. I cannot say much about it's complexity, but provided that my "dirty rectangle"-alike set operations perform in acceptable time (which I believe cause they require just a few instructions), it could get a nice O(n) performance in best case (no non-opac elements) opposed to O(n^2) in the worst case with only non-opac objects. But I guess that special case is worth it's own implementation without opac-centered drawing reductions.
However, I'm pleased with the current concept. If everything turns out well, I can profit from a simple and relatively easy to implement rendering. I already thought about possible uses, interface concepts, etc... Hopefully possible with this new technique. Provided that it works as quick as expected - which can't always be guaranteed if you don't test or prototype it.
Hooray! I gonna draw some mockups now.