> I feel like such a hybrid device would have a number of obvious operational advantages, e.g. much lower bandwidth/render-power requirements (each frame, your GPU would only have to render a high-resolution image of a little square, along with a very low-res image of the rest of the scene.)

Foveated rendering provides these benefits without needing a mechanical mechanism. (Though it does still require full-field high resolution displays and apparently the eye-tracking requirements haven't been fully solved yet -- which would also preclude your suggestion of a hybrid device.)