Considering the count rate limited nature of the FOC, it should be clear that achieving high, or even good, signal-to-noise, while at the same time maintaining acceptable levels of linearity, is rather difficult. Only the smaller formats will accept relatively high count rates and give good linearity response. Unfortunately the data is then ruined by edge effects.
The most important reason why S/N should be taken into account when considering image restoration is that it is the S/N in your data which determines the relative importance of having an accurate and applicable PSF to do the restoration with. If your data consists of faint, diffuse, extended source emission, with low S/N (peak counts of less than, say, 50 counts/pixel, there is no particular need for high S/N PSFs. The results obtained from the data using, for example, a narrow width Gaussian approximation for the PSF, would probably be as valid as if a perfect empirical FOC PSF had been used - it is the data itself which provides the limitation on image restoration. In any case, since the intrinsic image structure is usually not known a priori, in general it is the ``nicest'' reconstruction which is subsequently used.