but, basically, combinations of: 1) switching to a ReadWriteLock, 2) moving

invoke_runnable

onto a separate CpuPool (currently happens on the main thread), 3) invoking all Nodes on a CpuPool, rather than just IO nodes