... However, if we look at the differential signal (Voutp-Voutn), the influence of the CMFB vanishes. The differential signal seems perfect.
Comparing the single-end signals and the differential representation, we may conclude that the former are not well settled, whilst the latter is well settled. (Confusion??)
Actually, in my practise, opamp exhibiting this phenomenon still works well preliminarily by simulation (this conclusion has to be fully verified further) , however, I'm pretty worrying about the result of the final testing.