I'd always felt that the main issue with using USB is that there's no guaranteed packet delivery time (which does exist for e.g. firewire), which is kinda important for audio streaming.
Regarding the link above..."oh bloody hell not this crap again"...
First thing I want to have a go at is this statement: "While this view has long been debunked". Has it bollocks. By whom, and when, where's the evidence? Use your head; reject any and all unsupported claims.
There are two layers of Cyclic Redundancy Checks for all transmitted USB data; 5 bit for token packets and 16 bit for data packets. If the CRC checks pass, you can be pretty damn sure the data has been successfully transmitted and received. ACK and NAK handshaking and ERR reporting are all implemented. How do I know this? Because I've worked on USB driver stacks, some pretty horrible ones too (see below...)
If you have poor cabling yes you can have far too many retries, poor transmission rates, lousy bandwidth or even a complete lack of functionality. Most if not all problems, imho, are caused by badly written USB drivers (man I've seen some that would make your eyes water and children cry), badly implemented electronics, out-of-spec components etc. As long as the drivers are well written, the hardware has been well designed, the components and cabling are in good condition then it will work as expected. It's science. And logic.
If your driver is some broken-ass numpty-written pile of crap then ofc you are going to have problems. I don't believe in space magic, nor do I believe in magical bit massaging hardware which corrects USB data transmission problems.