It may be of interest that although Erlang has been doing lightweight concurrency for >20 years, - you can choose whether you want to use an SMP version that has as many schedulers as there are cores (plus internal locking as needed) or a non-SMP version with one scheduler (and no internal locking); both versions are standard and it's only a performance issue, not a semantics issue - performance sometimes goes one way, sometimes the other - there was a "one UNIX process per Erlang process" implementation; I have a copy. The community interest in it was, shall we say, massively underwhelming. It might also be interesting to note that the experimental operating system K42 from IBM does _all_ user-visible threading in user-land. This includes thread switching and even I/O blocking and unblocking; all done in user-land. I don't think we've begun to explore all the options yet.