It is an odd idea and I don't see it really working that well. I did some work with bulk MEMS structures my self, but never god the idea of using it as a voltage reference - more like as an VCO with a little more pull range than a VCXO.
A point that worries me in this application is the surface oxide layer. If very thin or no oxide one would have to expect some more growth of the oxide which would via surface stress cause large shifts. If there is a significant oxide already, the oxide can react to humidity, even tiny amounts, like a fraction of a surface layer. To this level even the TO8 can may not be sealed enough. At least from my experience with a possibly different oxide layer, the time to equilibrate with humidity is more like in the months to years at room temperature - so just the wrong time scale.