Use the same logic for returning the samples from the various MPI tasks
as for working with them in each task.