I wonder if would it be possible for nVidia to rewrite the code for the frame generation pipeline, so that it uses forward in time interpolation in VR only, the same as motion reprojection?
If so, it could possibly replace MR with a faster solution and better AI frame rendering.