The graphs suggest that instead of using extension and suffering from unsharp corners, we should try to keep the lens at its designed optimum and transform the resulting image by other means, either shrinking it with a Speedbooster or enlarging it with a good teleconverter. Some people think that either of these devices will degrade the image, but that is not necessarily true. At their optimum magnification, the Printing Nikkors may produce an aerial image that has more resolution than our sensors can record, so a mild teleconverter might be able to enlarge this image to show more detail than was visible in the original recorded image.
I have the PN105 A that was tested at coinimaging.com (thanks to Ray Parkhurst). The graphs show that its corner sharpness falls off very quickly when m is different from 1.0. This suggests that in order to get a sharp image at m> 1 it would be better to set the lens to its optimum m=1 and use a Nikon TC14 or TC17 to expand this image, rather than getting the higher m by extending the lens beyond its design point. Likewise, to get an image at m<1 it might be better to set the lens to m=1 and use a Speedbooster to compress this perfectly sharp image, rather than getting the lower m by reducing the lens extension far from its design point.
Here I test both approaches. I used a PEN F in high-resolution mode, which produced 50Mp jpgs on a tiny sensor. With this I can see differences that might not be easily visible on sensors with lower pixel pitch.
First, I compare the results of extension vs TC for m=1.4 (since I have the Nikon TC-14E-III teleconverter, whose stupid "child-proof" mounting tab I had to destroy by drilling through it).
Here is the 8160 x 6120 image at m=1.4 (at the reduced resolution of this forum, there is no difference between the TC image and the extension image). I have not done anything to it except reduce and compress it for the forum (too bad it needs to be compressed, it is stunningly sharp when viewed large on a monitor).
Here are 200% crops of the TC-14 image (left) and the image obtained by extending the lens to get the same magnification (right). (The images were so sharp, even in this interpolated high resolution mode, that it was helpful to look at the 200% crops instead of the usual 100% crops.) The PN105 is set to f/2.8 for the TC image and f/3.3 for the extended image, because the coinimaging.com tests showed that 2.8 was the sharpest aperture at m=1 and 3.3 was the sharpest aperture at 0.66 and presumably its conjugate. (I confirmed those results with my own tests: 2.8 produced the sharpest results with the TC, and 3.3 produced the sharpest results with extension):
Left: PN105 + TC14. Right: PN105 extended.
It is clear that getting to m=1.4 by putting the teleconverter on the lens, which was set to its optimum m of 1.0 (left), gives much sharper results than just extending the lens to get 1.4x. This should dispel the myth that putting more glass in an optical path is somehow always bad.
I also tried the Speedbooster Ultra to reduce the magnification of the PN105 when it was set to m=1, and compared this to reducing the extension to get the same image. Here is the image:
On the left is a 100% crop taken from near a corner of the Speedbooster image and on the right is a 100% crop of the image obtained by pure extension. The difference here is slight, but the Speedbooster image seems slightly better.
Left: PN105 + Speedbooster. Right: PN105 on reduced extension.
I think these techniques would also work on other highly-optimized special lenses such as the Rayfacts that are optimized for 2x. I suspect the results with the teleconverter would apply also to APS sensors and maybe even full-frame sensors, since the converters are designed for full frame. .
