Posted 03 December 2013 - 11:47 AM
This issue your seeing is the very reason Deep Compositing was invented by the guys at Weta Digital and Animal Logic.
As Tilt says in normal situations you only have one sample of data for each pixel, but with Deep data you have multiple, maybe 20 or 30. With this much information compositing based on depth works much better as the anti-aliasing is preserved.
I've been running some test recently and I'm getting good results out of RenderMan, Arnold and V-Ray.
Hope this helps, its a really interesting subject. Unfortunately the file size of Deep data is pretty big so many studios are not using it.