Why USB cable makes a difference? it is because no error corrections for audio usage?
The test was done using a DAC with a poorer USB implementation.
In fact the Schiit DACs were known for this and is why Schiit has other products to improve on that.
So the 'cable makes a substantial measurable difference' which may not have to be audible at all only applies to some DACs.
As mentioned by the OP this seems caused by the length of the USB cable.
One thing that is of importance is that this DAC is fed from the USB. This means that the power supply current to the DAC passes through the USB cable.
This is the most important factor in this case.
Long (generic) cables most likely have thin wires inside. This means there are losses that are not audio related but drawn power from digital (switching) circuits related.
So there will be some 'spikes' between ground (0V) and the 5V power supply.
This should not be a problem when input filtering of the power supply is sufficient.
This is theoretically very nice BUT there is also something like PCB design. The way wire (traces) run over a PCB that has BOTH digital signals AND analog signals.
Make a small groundplane or layout or groundplanes coupling error and some of the digital 'nasties' can actually go into the audible path.
This is even more true when there is also a groundloop with the amplifiers connected to the DAC.
Not saying this IS the reason, merely it could be one of the reasons.
So in this case the most likely scenario is that due to resistive losses in the USB cable a very small percentage of the current drawn from the digital circuits may have found its way in the analog path which is what is measured.
As can be seen we are talking -126dB peaks (relative to 2V) so we are in 1 microVolt territory with the long cable for audio or the digital receiver used is misinterpreting 'decision levels' and the rest of the DAC is more sensitive to the jitter this causes (poor jitter rejection).
Another issue could be (since this was shown in a jitter test) that the used USB receiver was not well implemented/designed.
Not so strange as Schiit just recently 'admitted' they finally fixed USB. One could read this as 'we finally made a good USB receiver and all that became before it was a bit flawed.
So the other likely scenario in this case is there was a higher than desirable amount of 'digital garbage' on the data/and or power supply lines (resistive and maybe even capacitive/impedance matching issues) that the used USB receiver had problems correctly deciphering the digital waveform it did receive and something in that circuit made jitter worse which wasn't properly reduced in the DAC itself.
So... we don't know for sure WHAT exactly caused this as it wasn't investigated.
The only thing that we can conclude is that in case of the Schiit modi2 DAC the length of a generic cable resulted in slightly higher measurement errors that may well have remained inaudible (skirting hasn't widened).
That is, when no digital volume control has been used at say -40dB or so.
This too we don't know as it has not been tested.
In short... it is unknown what caused the difference in measurements but seems related to a specific DAC and cable length (cable resistance/bandwidth related ?). It can be caused by a number of reasons.
It was also shown that another DAC was not 'affected' at all.