My OP asked about need not preference for expensive USB cables. This could have been posted more clearly. Need is about functionality. Preference is about whatever floats your boat including SQ, looks, durability, and reliability. But, I am glad that we are talking about both. Because if it comes down to bit errors (I think it does), what I first viewed as a preference may actually be a need. I can further elaborate on this by posing a question and then answering it.
If bits are either 1s or 0s, how can USB cables provide a gray scale in SQ?
Short answer:
Bit errors due to insufficient margins.
Long answer:
I believe digital audio has the misfortune of being a real-time application that uses designs intended for non-real-time applications. In real-time applications, there is one chance to get it right. Large margins are specified for all component designs in order to prevent bit errors. Typical USB applications (data transfer) are not real-time. They do not require large margins because bit errors are acceptable. You have the luxury to either retransmit until the received packet is error-free or you apply forward error correction to fix the error. We do not have this same luxury in audio. While music will flow with bit errors, those bit errors distort the analog wave shape (lower SQ).
Of course, isolation is also an issue. Al has an excellent post above that discusses isolation. But in this case, the cable solution is a bandage not the issue. Therefore, while I fully agree with you Al, I left this for another time.
My guess is that one USB cable may provide better SQ over another if it has better analog properties that prevent eye closure and transmit fewer errors. Having said that, what is good enough? Is it the cable that costs another $50 or $500? I think the cable industry intentionally does us a disservice of keeping us in the dark in order to sell into that ignorance. The test would be simple: Pick a nominal setup and test for data rate vs bit error rate. Then you pick an acceptable bit error rate (let's say your typical song is 4 minutes so you pick BER < 1 per 5 minutes) and select the least expensive cable that does not exceed that bit error rate.
