Originally posted by Breakfastographer I applaud your NR tests, but with respect I'd say the level of noise reduction used is rather heavy (for both cameras). IMHO, there's no need to remove all noise from a 100% pixel-peep image.
Take the following examples, processed in Darktable 2.4.4...
Here's a 100% crop of the same ISO 12,800 K-1II image you used, with AMaZE demosaicing and no noise reduction or sharpening applied:
... and here's the same crop of the K-1 image, same demosaic algorithm and no noise reduction or sharpening:
We can easily see that the K-1II image (the first one above) has less colour and luminance noise - exactly as we'd expect, given the baked-in NR.
Now, here's the K-1II image with some colour noise reduction (using Darktable's Equalizer and De-noise (Bilateral) tools), and a little gentle sharpening. Note that I haven't applied any luminance noise reduction... It's just not necessary, as this level of luminance noise isn't noticeable once you view at sensible reproduction sizes, or from sensible distances:
... and here's the K-1 image, also with colour noise reduction (but more of it was required), but here I had to add luminance noise reduction too (using Darktable's Equalizer tool) to more-or-less match the quality of the K-1II image. Sharpening level is the same, though:
As you can see, in this simple test I was able to match (fairly closely) the output of the K-1II by carefully applying stronger noise reduction to the K-1 image. But the most interesting outcome is that the detail in both images is largely the same. If there is a slight edge to the K-1, I can't see it. But either way, it's negligible, and - for this photo at least - won't be noticeable at typical reproduction sizes and viewing distances. You might just see a tiny difference when pixel-peeping at 100%+, but none of us actually views photos that way.
As a point of interest, I tried the same experiment using VNG4 as the demosaic algorithm. Here, the K-1II file did in fact appear to lose a
tiny amount of detail in the green feathery areas compared to the K-1... but it really was minimal. Not enough to matter in real use. This further reinforces just how important the processing software and user skills are in getting the most from either camera's files.