In screen mode for 18% SNR the D70 does better than the 645D.
And so it should be, since the Nikon D70 has larger pixels than the Pentax 645D. It is not as good as a camera, because it has a lot less pixels, but "screen mode" refers to the noise level per pixel.
To explain again why I compared the D610 and 645D: the question is what the change between ccd and cmos technology will bring for MF cameras. My idea is to compare the 645D, a ccd camera, with a hypothetical 645Dcmos, that is the same camera, same sensor size and same number of pixels but with a cmos sensor. That 645Dcmos should have a better low light response, but by how much?
To determine how much better the 645Dcmos could be, I check how much better the noise level per pixel
is between a recent ccd camera of a given pixel size and another camera having the same pixel size
but built with cmos technolgy. These two cameras are the 645D (again, but we don't have much choice as to recent ccd cameras) and the 610D. I find that the cmos pixel is about one stop less noisy than the ccd pixel. Thus, I conclude that the 645Dcmos, should it come to market, would be about one stop less noisy than the current 645D.
It is an approximation, but it should give a reasonable ballpark estimate. The end camera will probably have 50 mpix and not 40 and may have other improvements or not. Still: I would expect the cmos successor to the 645D and also the new IQ250 to be roughly one stop better than the current 645D (or H5D-40, which uses the same sensor as the 645D).
Compare that to the current press release of Phase One and a recent lula article which give the impression (granted: without actually saying it) that the cmos MF cameras are unusable beyond iso400 and that the IQ250 would be great at iso 6400...