Hi Edmund,
I am aware of MTF comparisons being tricky, but please note that I explicitly said "diffraction limited lenses". MTF drops almost linearly with increasing frequency on diffraction.
Post processing is of course a dominant factor, but the more of the MTF is coming natively from the lens the less we need to sharpen.
So very clearly, assuming a linear drop of MTF over frequency is an oversimplification.
Below are MTF plots measured on a Planar 80/2.8 at f/8 and a Sony 16-80/3.5-4.5 ZA at 80 mm and f/8 on 6.8 micron P45+ resp. 3.8 micron Sony Alpha 77 SLT. Near optical center and with no sharpening in Lightroom. Here the curves drop rather nicely. In this case the DSLR is an APS-C camera, so image circle is around 28 mm while on the P45 it is around 30 mm. So and image shot on the P45+ needs half of the magnification of the APS-C camera. So here we see that MTF at 30 lp/mm is around 0.64 on the APS-C, but for the P45+ we would check the curve at around 15 lp/mm and get an MTF of 0.82, clearly better.
The next figure is with a bit aggressive sharpening. Very clearly, sharpening can twist things around. If we still look at MTF at 30 lp/mm on the APS-C, we would end up at 0.92, clearly better than the unsharpened P45+. Looking at the P45+ at 15 lp/mm it would have an MTF (after sharpening) of 1.01. This would give a visible difference to the advantage of the P45+. Very clearly, once we apply sharpening things get tricky. But, everything kept constant a larger format should offer an advantage, until we run into the diffraction limit due to DoF requirements.
Erik, I'm not so sure you're right. MTF 50 is an accepted single-number way of measuring, but you cannot scale the MTFs around so easily as the curve is not the same on all lenses. By analogy with circuits, some lenses will drop at I guess 6db per octave or whatever at the frequency you are at, some are still flat etc. Also, MTF gives you the native (hopefully) (Imax-Imin)/(Imax+Imin) at that spatial frequency, but afterwards you can postprocess ...
Mixing reasoning in linear spatial and Fourier db spaces is not always a good idea, at least it seems counterintutive to me with my very rudimentary understanding of mathematics
Edmund