I think you mean oversampled ie there are more than 2 pixels per star FWHM. (Why do you consider 1.7 arcsec pixel optimum? What is your actual best seeing?)
Undersampling is to be avoided for science applications but the effect of the additional read noise on signal/noise ratio due to oversampling (this is the only disadvantage I know of) depends on the observation. Playing with Michael Richmond’s calculator for example entering actual numbers for different cameras and observing conditions will give you an idea of the significance for photometry.
Note with CCD cameras (not CMOS), in camera binning (as opposed to post binning) reduces the read noise as there is only one dose per binned super- pixel, though CMOS cameras generally have lower read noise