Actually, it makes no sense to use a CoC smaller than the lens/sensor system can resolve. The overall resolution O is:
1/O = 1/L +1/S
where L and S are the resolutions of the lens and sensor, respectively.
[a href=\"index.php?act=findpost&pid=101394\"][{POST_SNAPBACK}][/a]
I think you make the "brick wall" mistake, of assuming that as son as the CoC is smaller than the resolution length scale 1/O = 1/L + 1/S, it has no effect on overall resolution. This is a version of the false belief that the resolution limit set by lens and sensor is the minimum of L and S.
It seems to me instead that no matter how small the CoC, it can smear some light that "should" fall near the edge of one photo-site over onto an adjoining photo-site, and so to some extent reducing resolution. (Likewise for even the very small diffraction spot size of very large apertures.)
That 1/L + 1/S is an empirical approximation, with 1/L and 1/S both in units of length, so says that the size scale of the smallest resolvable feature is roughly the sum of the size scales due to each factor, lens and sensor. That is, angular resolutions combine additively.
This might generalize to include OOF effects at a particular location in the image as something like
overall resolution length scale = Ab + Pix + CoC + Dif
where CoC is the diameter of the circle of confusion at that location in the image, Dif is the diffraction spot size, Pix is pixel size and Ab is the resolution size limit set by lens aberrations.
This indicates that overall resolution is improved somewhat by reducing any of the length scales (reducing pixel size, CoC size or diffraction spot size, or increasing lens resolution) even one that is already smaller (better) than the others. Of course the greatest benefit comes from improving the worst factor.