So the big problem with zdepth and flaws in post dof is that a normal image can only save pixel information for each of the objects that the camera sees, it can’t save anything for objects that are in behind those objects so when you start using dof, the blurring on close objects can almost let you see in behind them slightly and makes the effect fall apart. The other second big issue is that a depth channel is only saving a single value per pixel, and if you have an anti aliased edge between two objects, that pixel is half one object, and half the other object. The Depth channel will only save a value for the object closer to camera, so you’re missing data for the further away object so the depth blur effect gets issues.
As deex mentions, one way to make this a bit better is to use deep compositing which WILL let you save information for objects in behind the objects closest to camera, including depth information. They’re much heavier files and you’ll have to use a compositing software that supports deep images like nuke or fusion though.