FEniCS + MPI on docker inefficient?

Maybe this can be the reason for bad performance?