pamilrts machine layer is less performant than pami machine layer

[evan-charmworks]

Original issue: https://charm.cs.illinois.edu/redmine/issues/2040

---

It has been noted that the non-LRTS pami layer runs faster than pamilrts, which is why we are keeping it around instead of deprecating and removing it. We should perform new benchmarks of both layers on BG/Q and ppc64le and work to resolve the discrepancy if still present.