Yes, thanks Wim, I understand your logic.
What we see, however is that in moving from 10MP to 16MP the centre performance improves but the edges/corners do not. It's not a small difference. It's not like moving from DX to FX as there is no physical effect from sensor size and a deterioration in the telecentric efficiency of the lens. It does in fact LOOK like a comparison of DX and FX! ? ? ?
I am also surprised by the test result of the 14-24mm on D7000. Althoguh I understand that as the sensor resolution increases (but the sensor size does not change) the system border/corner resolution becomes worse relative to the center resolution (at the same aperture and focal length) since at higher sensor resolution the system resolution resembles more the lens resolution, the border/corner relative to center resolution is worse on D7000 (DX) than the same lens on D3x (FX). Is this related to the magic "12-14MP" or "16-18MP" mentioned by Wim?
This is exactly the effect I am talking about. If you look at tests done on film, you'll find that the corners are always lagging behind the centre, resolution wise. The curve for the corners generally follows the same form. but just lies lower in the graph. Occasionally it tails off more, especially wide open, and occasionally they get very close together at some point.
With digital, if you look at some of the older reviews here, with, e.g., 6 and 8 MP cameras, you will find two things:
1) the edge and centre curve are often quite close together
2) there seems to be a clear flat top to the curve at apertures from, let's say, F/2.8 to F/8
Also, when you do the maths, you'll find that the difference between system resolution at MTF-50 and the Rayleigh criterion, which from a lens diffraction POV is more than a factor 2 different, in system resolution results in only a few line pairs per mm difference in this aperture range.
With the newer cameras with higher MP-counts, and a good example is the 5D II, where one sees that the curves now form more of a parabola rather than a rounded block wave curve, and that often edges start at much lower resolution. However, the resolution is still above analog levels.
Coming back to point 1), I think that is caused by the fact that a sensor doesn't have the emulsion effect of film, which enhances a parabolic curve, and that the resolution is still relatively low and of little influence, relatively speaking, to system resolution. When you do the maths, you'll find that there is really only a few line pairs per mm difference at lower sensor resolutions for relatively high lens resolution figures.
Point 2 is caused, IMO, by the effect of a sharp limit to resolution of a sensor, essentially because of zero thickness of the sensor (no emulsion effect) and because of the AA-filter, which acts as a cut-off filter.
Of course, at apertures of F/5.6 and larger, optical aberrations tend to play a very large role, which is in principle also an equalizer for apertures up to about F/5.6 quite often, except for the very best lenses out there, and with APS-C bodies, because they grab the sharpest part of the image, the fall-off is less with those. Hence the reason why with APS-C bodies this seems to happen only at relatively high MP-counts vs FF, which records the extreme corners of a lens formed image after all.
Finally, on some review sites, this is more clear than it is here, not only due to the different way in which test results are obtained, but also because of processing and number crunching differences. If you have a look at traumflieger.de or lenstip.com, you will find resolutions which actually relate to system resolution at MTF-50 rather than the numbers you will find here on PZ, where they are crunched to get the absolute maximum out of each lens at each point. However, for photographic use one really needs Rayleigh criterion numbers, but that is something no review site currently provides, unfortunately.
Kind regards, Wim