An important point to realize is that shuffling the charge backwards and forwards on the chip can be done thousands of times faster than reading out information stored on the chip through the amplifiers. However, this cannot be done indefinitely due to charge transfer inefficiency and trapping sites that begin to build up in the silicon. I wrote short papers in 1994 and 1995 showing that you could get away with a hundred shuffles before these effects began to hurt you.
Another key point is that the sky emission from the atmosphere varies both spatially and temporally. I determined that the optimal switching rate is about 30 sec to 1 min, and the nodding distance should probably not exceed 10 arcmin. In the case of multi-object spectroscopy, you need only move a few arcsecs since light is only collected through small slits, rather than from the full field of view.