Coming back to where this discussion started, originally I wanted to make a point the few colleagues who stated rather categorically that "USB cables must all sound the same since they only carry bits", and the fact that they don't only do that (electrical noise aside, they also carry power for USB powered DACs), and if different cables are able to isolate the effects of high currents passing through the power cable from the signal ones, then not all cables sound the same in all circumstances if these currents can have a deletereous effect on the signal (timing at the receiver, increased noise, etc).
USB 1.0 and 2.0 cables are made of 4 unshielded wires. The 4 wires are: GND, +5, and two data wires operated as a
differential pair. USB 3.0 adds two additional
differential pairs (internally individually shielded or using twin-ax cable). The point of the differential data pairs is noise immunity. The device that receives the data looks for a difference between each wire of the pair and ignores any signal that is common. A noise signal from an external source that is coupled to the wires, either inductively, capacitively, or electromagnetically arrives on all wires equally, making that noise signal common mode, and easily rejected by the use of differential data signals. Noise that may be on the Gnd or +5 wires that may couple to the data pairs is also coupled equally to all wires, also making that signal common mode, and easily rejected. When an overall shield is added to some cables (not part of the USB 1 or 2 spec), and grounded, it provides electrostatic and electromagnetic shielding against external signals, reducing coupling, except for the inductive component for which the shield is useless. However, should an external noise signal be inductively coupled, or manage to penetrate the shield since no shield is perfect, the coupling would still be common mode to the data pairs. USB 3/Superspeed cables have by specification their data pairs internally shielded to maximize noise immunity and minimize data crosstalk.
When a differential pair is used and applied to a differential receiver, the common mode rejection ratio should be at least 50dB, but 70dB is not atypical. Assuming worst case of 50dB, that means that if you could (and you can't!) induce a noise voltage into a USB cable equal in amplitude to the 5V peak data signal, common mode rejection would see it as a .016V noise signal. However, considering the actual strength of an external noise source and coupling losses, induced noise voltages are far, far below that. The integrity of the data is ensured by the differential pair technique.
The single biggest problem with USB of all flavors is that a ground is carried end to end, which presents the strong possibility of a ground loop if the devices being connected have a difference in ground potential. That difference in ground potential is a problem, a design flaw, that can be remedied in other ways. But because when USB cables include an overall shield, and that shield is grounded at both ends, the resistance of the device to device ground connection is reduced, and both devices are brought closer to the same ground potential. This may in some circumstances reduce ground loop noise. But it doesn't affect the data because it's already immune, being differential. Again, ground loop noise is the result of a device design flaw, or a flaw in installation, it's not supposed to be there in the first place, and could be remedied without the use of a special USB cable.
The USB interface was designed to transmit data between devices without corrupting the data. That's why differential data pairs are used. The chance that actual data corruption is occurring because of coupled noise is extremely unlikely.
When audiophiles "listen to USB cables", and hear things they describe as more "open", or "better detail", these are not qualities that can be achieved by altering the data path. Problems in the data path only result in corruption of data, which only can result in increased noise. Any of the changes described by illustrative or analogous terminology could only occur if data were altered via some form of DSP that could affect a change without corruption.
One last thing, the capability to provide bus power over USB has a number of well defined specifications and different levels of power/current. However, the thing to note here is that devices must tolerate a 5% voltage fluctuation, and in the case of USB 3, operate down to 4V. That implies power conditioning and regulation at the device, making that device immune to noise on the power wire. This doesn''t eliminate the ground loop problem, but does reduce the possibility that noise induced into the 5V line would inject itself into the device that way.
