>That is why I was careful to state that the dynamic range can be improved by a factor of NxN for N-fold post-binning, and not the signal to noise ratio. Sometimes the dynamic range is particularly important, such as when trying to detect extremely low contrast objects for instance.
Indeed and this is a limitation of in camera binning where unlike post binning, you cannot see the individual pixel counts, so to be sure of not having any individual saturated pixels you have to limit the total counts in a super-pixel to the full well depth of an individual pixel.(Well I suppose one could perhaps assume a gaussian distribution of brightnesses in the star image and allow a higher maximum count per super-pixel which would bring some dynamic range benefit.)
The in camera binned exposure could be split into NxN shorter sub-exposures to match the post binned dynamic range without risking saturation but that then returns you to the same read noise contribution as for post binning. In camera binning can be advantageous though for weak signals in a dark background where the read noise contribution can be significant (in spectroscopy at low SNR for example.)