reduceExposure.py doesn't parallelise over all the input dataIds, although the processing is initially independent.  Please make it possible to run all visits/spectrographs/arms in parallel at least as far as a possible sequence point at full-focal-plane sky subtraction.  As I believe that there is not yet a need for such sequence points, it would be reasonable to delay that scatter/gather work until it's needed (and then probably we'll do it in gen3 middleware anyway)