So just for fun, I went back to the P to P site and plugged in the D500 against the D850 and the D850 DX. That produced even more interesting results (at least to me). Specifically, despite having slightly smaller sensor elements and a little bit older sensor technology (and no BSI design), the D500 shows as being better than the D850 in DX mode by about 0.3 stop, all the way down the line. I am personally convinced that the D500 is quite a bit better than most on this forum give it credit for being, but still believe that this is also a surprising result...perhaps even more surprising to me than your initial discovery.

I'm not too disturbed by any of this, because I firmly believe that actual performance and experience trumps any test results claimed by any single source. But I do think that an inquiry to Photons to Photos asking why they are displaying these at least apparently inconsistent results is not out of order.

Noise increases because of the different total areas of the sensor used. Use the whole sensor and you record more total light than if you use the smaller cropped area. Noise will increase/decrease with the total amount of light recorded. SNR is a ratio and goes up or down with the total amount of light recorded.

Try this analogy: You have two cookie pans -- one is 11 x 15 inches and the other is 9 x 12 inches. You put them both out in the rain together and allow them to collect water for 1 minute. That 1 minute is exposure time. After a minute it stops raining and you measure the depth of water in each pan -- it's the same. That's exposure. But now pour the water from both pans into two separate bowls. Do both bowls have the same amount of water? The larger pan collected more water. In the same way a larger sensor will collect more light. Noise (shot noise which is the noise we're concerned with) as a ratio to the signal will decrease with more light collected -- stronger signal.

All else the same, bigger sensors are less noisy than smaller sensors. Noise limits DR.