If the attribute you use to split traffic is different from the attribute you use to analyze results, it can potentially cause data quality issues since it could look like some users saw multiple variations of an experiment. In this particular case, I'm guessing the device id and Mixpanel's distinct_id are probably very highly correlated (it would be unusual for a Mixpanel id to change, without the device id also changing). So it should be fine.