I would have thought it actually helps the branch target predictor rather than the branch predictor. If you assume a simple predictor where the predicted target is just the last taken one then it's going to be wrong almost every time for a single switch. It will only be right for repeats of the exact same instruction.

If you have a separate switch at the end of each instruction then it will be right any time an instruction is followed by the same instruction as last time, which can probably happen quite a lot for short loops.