Bart van der Wolf, who seems very knowledgeable in the area seems to estimate the improvement in resolution to about 6%.
Hi Erik,
That number is used a bit out of it's original context. Many moons ago, I've done a
simple test which determined
the effect of demosaicing on Luminance resolution. With the demosaicing algorithm used, there was indeed only a loss of resolution of 6.4%. In addition there are specific color combinations that pose worst case scenarios for demosaicing of Bayer CFAs, compared to situations where there is a luminance difference between sensels. While significant, the demosaicing loss is not the end of the world. Slight defocus, or some diffraction blur, or camera shake, or heaven forbid a combination, does more damage.
A much larger difference is caused by the use of an AA-filter on the majority of 35mm or cropped Bayer CFA sensor array cameras. However, on most MF backs that is not a factor of concern. In addition, as you said, there are differences in MTF due to magnification factor and lenses.
Furthermore, having a lot more sensels, i.e. independent spatially discrete samples, opens up a lot of resolution. Having 40 M Luminance (in a CFA arrangement) samples is something more useful than 15 M RGB samples for the same field of view, even if the chromatic resolution is somewhat limited compared to the luminance resolution.
Cheers,
Bart