You are both correct. The thing to note is that the current version of mental ray handles motion blur with the option of key frame at start of motion, end and middle. You can output two sets of images (one for start and the other for end) to complete the set for Blender and sock them in with a combine node. The Z remains the same for both images in the set, so you can connect the Z of one and it’ll do it. This is theory at this point. I’ll have to experiment and see if it’ll do the trick.
If there’s a way to successfully separate this data from a single pass for both before and after frame data, then I’ll post it. Or if you guys figure it out, beat me to it, by all means.