This is at least partially related to the mx_stream issue I reported previously. I restarted mx_stream on c1iscex and that cleared up the models on that machine.
Something else is happening with c1sus. Restarting mx_stream on c1sus didn't help. I'll try to fix it when I get over there later.
I managed to recover c1sus. It required stopping all the models, and the restarting them one-by-one:
$ rtcds stop all # <-- this does the right to stop all the models with the IOP stopped last, so they will all unload properly.
$ rtcds start iop
$ rtcds start c1sus c1mcs c1rfm
I have no idea why the c1sus models got wedged, or why restarting them in this way fixed the issue.