Hmm I thought the depth blur node DID use the Z info. It is just a post effect, so any close (infocus)edges that are next to very distant edges will be unavoidably blurry. To elliminate this you need to render background and foreground elements seperately then composite them.
I don’t see any problems with…it’s all in learning how to use it. ALL Blender composites are better off with FSA. Besides, it beats the alternative of down-sampling Hollywood sized renders and you get a myriad of sample filter options V.S. some single, crummy, scalar, down-sample which many programs really suck at.