The danger of relying on tests that are not in controlled conditions is that you can make the wrong assumptions about how things like noise compare.
I'm assuming that the lighting was the same for these images. Yet, exposure is quite different.
For example, the ISO 1600 image from the K10 was exposed at f/5.6 and 1/25 second.
That's a full stop brighter compared to the ISO 1600 image from the Canon (which was exposed at f/8 and 1/25 second), and noise is going to be worse in an underexposed image. f/5.6 is exactly twice as bright as f/8
It was also exposed brighter compared to the other camera models. For example, there's about a 2/3 stop difference with the D80 image (it's ISO 1600 image was exposed at f/5.3 and 1/40 second, compared to f/5.6 and 1/25 second for the Pentax).
I see the same thing with the Sony Alpha tests. It used the dimmest exposure of all for it's ISO 1600 image (1/60 second at f/5.6).
So, what you're really comparing is metering and ISO sensitivity, not noise levels, since the tests were not using controlled conditions.
Heck, even if you were comparing the ISO 800 image from the Sony to the ISO 1600 from the Pentax, the Pentax would still have the advantage. The Sony was using a faster shutter speed shooting at ISO 800 and f/5.6 compared to the Pentax K10's ISO 1600 image at f/5.6.
For that matter, the Sony was using shutter speeds almost as fast for it's ISO 400 image at f/5.6 as the Pentax was using for it's ISO 1600 image at f/5.6
If you don't expose the images the same way, you can't judge noise levels. Even then, you'd need to find a way to adjust exposure for ISO speed sensitivity (the Sony and Canon models are about 1/3 stop more sensitive than their rated ISO speed, but the exposure differences in this test were much greater than that).