How many frames in a stack for good outlier rejection performance?

One reason to take N frames of a subject, rather than one long exposure is that during stacking it is easier to get rid of anomalous things like airplane trails.

With N frames, simple averaging would drop the intensity of an anomaly by 1/N.  That is nice, but outlier detection in the stacking algorithm - like kappa-sigma or others can improve on that.

But it obviously takes some number of frames for this to be effective.  In the extreme case of N=2, the outlier detection can't really work very well.  Even if it did pick the right frame to throw away, you would be throwing away half of your exposure.

My question is what is the minimum value of N where the stacking N frames with outlier detection in APP do a good job?   

My guess is that it is on the order of N=10, but I am not sure.

The cost of N becoming large is read noise, but many modern CMOS cameras have very low read noise - so low that sky background dominates it in many cases.