Since the read operations of HDF5 are not concurrent even with the threadsafe version, we have to use a distributed WorkerPool to speed up data loading.