I don't know if this helps, but my strong suspicion is that "None" is not handled as "do not override the expression", but rather that the expression is always overridden (which the "switch to plate, watch expression change" behavior seems to support), and if you do "None" it picks the next expression keyframe in the pose animation to use as the override. Which kinda makes sense; it's not how I would have done it, but it could actually be a reasonable abstraction to keep bits of logic separate. But it feels like it only jumps forward to "next frame", never backwards (e.g. "nearest frame")... and so presumably if the animation relies on a "back to default" blend to lose the expression as the animation ends, you might have basically no expression keyframes in the later part of the pose. And thus, presumably, get the "straight face" pose as a default.
Again, just a gut feeling, but I've had better luck freezing pose expressions earlier in the animation rather than later, which does at least provide circumstantial evidence.