A notable thing about these results are the very best results are when you have a one device clocking another device. You would think loopbacks would clock even better. Same clock used for both ADC and DAC. However, there is a small time delay caused by the signal leaving the DAC, traveling over the cable, and getting to the ADC. A few nanoseconds. You get lesser results if you have a master clock clocking both devices, and I think for that same reason. If the DAC clocks the ADC, then the clock signal also travels over a cable between devices and the clock between the DAC and ADC can be much closer to the same timing on the signal at both ends. Which means the nulling software has a much easier job time aligning everything.
I suspect the Diffmaker software limits how well it can match up the timing. So some devices in pure loopbacks probably are actually doing better, but the offset timing interferes with results.
And anyone interested in this, you really owe it to yourselves to get Pkane's Deltawave software. It does what Diffmaker does, only better, without blowing up all the time and with many useful features. You can get it for free here (Windows only):
https://deltaw.org/
It is up to revision 45.
PS-I've submitted a few results to that gearslutz thread, and Deltawave gave results within a db or 2 of Diffmaker. More recent versions usually give better results by just a bit. Plus Deltawave is so much more nicely featured.