This story might also overlap to aggregations of files stored on the POSIX file system as well.
Use NcML to time aggregate files in the cloudy dap data collection.
Faster than 'regular NcML.' Likely explanation: the dmrpp files relieve the handler from the task of building dataset metadata for each file. The MDS would do this but not for HDF5 because the DDS/DMR objects used for the responses are not the same as the ones used internally by the handler. The h5 handler uses specializations of those objects.