Sample motion blur make more sens in cycles then in BI. The render samples can be distributed over the motion blur samples (if every motion blur sample is given it’s own seed). So a 4000 samples render with 200 motion blurs samples will render 20 samples for each motion blur sample.
We still need to build the BHT for each motion blur sample - but the actual render need not cost any extra (in contrast to BI where a sufficient number of motion blur samples get insanely expensive).
With sufficient motion blur samples that is the best method, think reflections in a fast moving train, motion blur on shadows, moving object passing behind each other, or any other indirect effect that just works with motion blur samples but need hard extra work with other methods.