I generally agree with Paul here. It's perhaps also worth pointing out that one of the main points of the csmt mechanism in the first place was serialisation of D3D commands. One could perhaps argue that that's unlikely to be an issue for applications limited to running on a single core, but it doesn't seem like a very robust argument.
Something worth exploring might be to run the CS thread as a Unix thread; I think those may not be affected by the affinity mask set for Windows processes. That's not quite trivial, of course.
Separate from the merits of the general approach, I don't think we should override a csmt setting explicitly set by the user. If we're going to do this kind of thing at all, it would need to be before applying settings from the registry and/or environment.