Multi-threading on Windows leads to segmentation fault for test models.

As I brought up on the email list.

Edit: It is probably better described as random behavior, since the same test model does not always segfault. I am presently running a test model with R under gdb to get a better sense of what's happening.