This guy's measurements look completely OK to me as I see the same things all the time in my tests when the hookup is not 100% error-free.
The point is, he's *not* actually measuring the noise at the
outputs of the devices, rather he dominantly**) measures the noise at the
end of the unbalanced cables at the analyzer's inputs. While he uses a separate ground lead in an attempt to divert the main leakage ("ground loop") currents away from the interconnects this is not sufficient, obviously. So what he is mainly seeing is the voltage drop from the balancing currents that develops along the shield of the interconnects. Simple 2-prong SMPS wall-warts are notoriously bad wrt these mains leakage currents.
**) However, some devices have such a bad internal grounding layout that any leakage current through the device already creates these voltage drop errors.
But hey, in some sense this is a very valid real-life scenario, of course, as that is what will happen quite often with unbalanced interconnects depending on the total system's grounding and supply schemes. The good specs of any piece of equipment then usually fly right out of the window as the mains leakage current "ground loop" noise will dominate SINAD etc. We (
@pma, myself and some others) have pointed that out already, over and over.
If one were to measure the true noise as actually present right at the device outputs one
must use balanced interconnects to the analyzer even when hooked to unbalanced outputs. By this, the cable influence is eliminated but any mentioned bad layout-induced errors will be seen. To double check, on needs a completely floating analyzer... then even the bad layout type of error is removed (this is important for example for headphone amps where the load is truly floating, not connected to anything else), and in the complementary test, defined leakage current through the device can be injected to get reproducible data.