Hnefi
Basically, when starting the game, spawn X threads and partition the world between them so that only one thread can write to any particular world block. Then, when calculating a new frame, make a read-only copy of the previous world frame. Any reads are done from that copy; any writes obviously occur only in the blocks assigned to a particular thread. Wait for all threads to complete, then render thread. No messaging, locking, event handling or whatever needed. Just care and forethought.
But you are right that Toady likely not possessing that kind of knowledge. That's the very reason why people would like to see others involved for things like optimization and UI design and the main point these people are trying to make; Toady seems either not able or willing to do it himself, and it could improve the state of the game tremendously.
I don't see what that has to do with using parallel programming to improve the performance of DF's simulation. If you are implying that threading implies nondeterministic behaviour in a simulation, you are mistaken. In fact, a system with such a discrete space model as DF would be quite easy to parallelize in a safe and deterministic way, with the proper expertise and forethought.