This is exactly what has been lacking all this time. We need a "game engine kernel" that is suitable for all types of games inside one genre. That way work has not to be duplicated across the plethora of projects that are out there, and collaboratively working on one Linux FPS engine will bring forth a new paradigm of how to design game engines. Just as Linux brought a new paradigm in OS design.
The best of the opensource tools put together, and working in a quite elegant way :-)
PS: 0MQ handles in-process communication using shared memory, so, pretty much no overhead :-)
Actually, 0MQ seems particularily well suited, as it doesn't try to manage reliability and in a game you don't want a queue of old input messages creating artificial lag. And I guess for sound, if you run out of resources to play sound effects, you can reliably skip that message too :-)
And a connector to AI as a multiplayer unit in another thread is much the same logic as an actual multiplayer game, since 0MQ doesn't care.
Hmmm.... I'm gonna watch this :-)