End product too expensive, leading to bankruptcy and personal humiliation

Zero g bias drift can be measured over temperature and compensated for by the microcontroller

Temperature induced effects are relatively linear so they are easy to handle using either a look-up table or a mathematical solution

I.e. add a temperature controlled gain stage with an op-amp and a thermistor

Hard to match response for sensor to sensor variance

Holds the accelerometer at a constant temp, so drift due to temp doesn't exist

Crystal oven may consume a lot of power (500 mA)

Let the user hit a "reset button" from time to time when the accelerometer is experiencing zero g

Does your application really need to be DC coupled (i.e. do you need to measure tilt)? If not you should AC couple the accelerometer, via a capacitor, to your signal conditioning stage to eliminate bias drift

Use software to detect very low speed, long term, zero changes