What I find really amusing is that back when USB was proposed I argued that rather than create an entirely new infrastructure why not just use Ethernet packets. There are additional protocol types available in the bit encoding and all sorts of infrastructure that already supported layer 2 switching. Plus the "phy" (10/100 at the time) was dirt cheap and already available. It never went anywhere, and at the time I figured that nobody wants to support a standard that they can't see how to make licensing money off of it. Somewhere I have VHDL code for a PC parallel port emulator that ran over my protocol. It fit in a small FPGA with the ethernet MAC so it wasn't like we were talking about a lot of gates. Just no consortium fees and no way to keep people out of the ecosystem you didn't like.
Doing USB over IP is an interesting way of re-creating that work (although mine was multi-master so technically a bit better :-)). Of course these days USB 3.0 is faster than the ethernet port on most machines so it was no doubt doomed from the start. But its fun to do. The SATA over Ethernet stuff that CoRAID did was like that, very nice work.