When a thread is rescheduled, the executor ID is left
set so the thread cannot run until the executor it was
running on comes back up. Presumably this executorID
should be blanked out so it can run on the next
available executor.
Reported on the alchemi-developers mailing list by
Reece Arnott.