Extending on Jerry's point (3)...
The statistical approach is good if you have the time and resources. YOU set the calibration frequency, and the driving force, I think, is the requirement that you justify all prior results if you find a gage out of calibration. If you set the calibration frequency too close, you'll never have a situation to react to, but you may be spending too much on calibration. If you set the calibration frequency out too far, you'll likely have an out of calibration situation that will be difficult and time consuming to justify (and may have a recall situation as well). Look at the history of the gage, not just whether it's "OK" or "NOT OK", but the actual found values. Set the calibration frequency based on the expected drift of the gage, so that you calibrate befor it drifts to the point you have to react.
I suggest (for small companies with limited resources) that they start out with a high frequency, then evaluate the situation after three cal cycles. If the gage doesn't drift, extend the cal cycle. Repeat. When you see a change, recalibrate and hold the cycle. Monitor for three more cycles and see if it works for you. This may not be statistically accurate, but it works from a logical standpoint. And in my mind, statistics (properly applied) is nothing but logic!