But doesn't that depend entirely on the synchronization mechanisms and how safe they are? For example, contrary to what you suggest I'd say that Go's stateless

But doesn't that depend entirely on the synchronization mechanisms and how safe they are? For example, contrary to what you suggest I'd say that Go's stateless green threads are hard to debug (though there are good tools) and easily lead to deadlocks. That's because the available synchronization mechanisms finitely buffered channels, non-re-entrant mutexes, and a few atomic operations are hard to use correctly. Higher level constructs like actors or transactional memory are in my opinion less error prone. The point is that e.g. futures are just some threads with a global synchronization mechanism for obtaining the result. Whatever makes the future stall will also make a low-level thread + your own synchronization stall. Or do you mean some more advanced failure-tolerant threading like in Erlang as compared to less advanced threading primitives like futures?