Power and data running side by side is only a big deal if the power line has noise. USB card with low noise 5V this wont be an issue.
I have tried Lightspeed 10G. Very sterile and lacking in nuance level detail. Beaten by even the series 6 Platinum. The full linear HDPlex PSU is the best fully ATX compatible solution I have tried. JCAT is better than PPA V2.
Galvanic isolation is necessary for Ethernet because it needs to connect between rooms which might have different ground reference and therefore might create ground loop. You can galvanically isolate USB if you like - it doesn't really help a whole lot and in my experience is a step backwards. Even isolating the ground pin is a step backwards. The galvanic isolation does provide some noise rejection and DC isolation but it also adds jitter. The net gain is just not there. Most of the good USB solution use galvanic isolation after the USB receiver, that way the PHY noise is isolated from the output. Again the net gain is not a given.
What makes you think Ethernet is free from PHY noise? In my experience NIC's put out a lot of noise. I need to enable my NIC because I am running a headless machine. If were not running a headless machine, I would disable the NIC as it produces much noise.
Secondly, have you measured the PHY noise and compared it to Ethernet, or are we just comparing one D/D implementation to another. There are so many more factors that might be affecting the sound of the Rednet vs your USB based D/D. Not least the quality of the reclocking, SPDIF output quality, power supplies etc. To simply decide that the difference is 100% down to using AOIP is an assumption. It could be valid assumption, but we would need to isolate so many other factors to make that determination.
I cannot account for people preferring AES/SPDIF over USB. As mentioned the interface is a bottleneck. If they were using LVDSI2S I would have slightly less reservations. Impressions come from so many different setups - perhaps they have poor computer sources. Also different systems tend to show or hide things in different ways. Even in my own system as it develops, positions are overturned. As resolution improved the true nature of things come out more. What was previously observed to by detailed turns out just to be thin or harsh. Could be that my system is still at an interstitial stage where USB appears to sound better. However my own preference for USB came only AFTER upgrading to the HDPlex PSU and case. We have to rationalise the world with the information we have available. All I know is that on a technical level SPDIF is a sure bottleneck, USB evils are contingent and for the most part unverified. LVDSI2S is much less bottleneck.
I used to use an external D/D until I improved my computer, but of course [computer upgrading] was costly and not possible for everyone.