Comparing the K3 to the D800E, using the whole frame the D800E is an improvement in noise both pixel to pixel and normalized according to DXO. But as we know, we are dealing with a totally different field of view, and the photo will be different.
The DXO screen graph essentially is a one pixel to one pixel signal to noise result, the print graph normalizes or tries to come up with a number that represents what the image will look like printed to a set screen or print size. It doesn't capture all the subtleties of the sensor, but gives an indication. Our photos generally are not 18% grey scale, or we hope.
I wanted to extrapolate to find what happens when you crop the image.
For example, yesterday an eagle was eating a fish on a pole beside the road. The whole thing happened in the morning, light was poor. I used the truck as a blind, handheld through an open window. 1/320, iso 1280, 500mm f4.5, right on the edge of capability for the k3. Cropped down to 3114x2489 pixels, or .31 of the total available.
On the D800E, in that situation, normalized, what would happen? The chart at the bottom shows them normalized curve at % crop of the full frame sensor compared to the K3.
Here is the cropped and finished image.
This image wouldn't be printed, it isn't good enough. I'm surprised I got anything, handheld at 1/320 from a vehicle usually ends up with a blurry mess. At 1600 iso it would have been worse, the noise was surprisingly not bad. The K3 seems to have a dramatic falloff point, and if you stay inside it is reasonable, but past it is garbage. Essentially to get to the same place with the full frame 32mp sensor I would need to crop 20% less; longer lens, be closer.
Everyone shoots different things, with different lenses in different circumstances. This is purely a selfish exploration, trying to figure out how this technology would apply to me.
I used Google sheets for the calculations and graphing, it is quite simple. If someone wants to play with this, let me know.