I'm guessing TBB would be the limitation, but to be sure, would it be possible to process R's long vectors in parallel? Since the APIs use size_t in the parallel for-loop, I guess that would not be handled correctly.