That seems like a problem of matching the workload to the hardware- graph computing isn’t embarrassingly parallel in the regular sense.