It's a shame that USB went with Vbus/Gnd rather than V+/V-. The latter would allow the negative side to drift away from ground (which is typically kept on the shield in a zero-current path). They had to devote a small section to voltage drops it in the latest spec (184.108.40.206 in USB_PS_R2_0 V1.1: bottom line is 375mV of Gnd rise and 625mV of Vbus drop).
The Ethernet guys realized this would be a problem so they made their signals transformer isolated. But USB is designed for short runs and cheap interfaces, not 100m runs across different electrical grids.
Note, however, that this case was not related to the ground potential at all. Vbus and Gnd were reversed, exceeding the Vgnd_drop limit by over 9V (2400%), or possibly 40V (>10000%) if the device had managed to negotiate the new voltage limits before it died.