There was a bug I carried since long ago, It was hard to detect and its reproduction was depending very much on how well the Slab allocator worked for net_buffers. The debugging shown like trying to send a buffer that was already sent... all fake as it was a reused one.
There were important internal features to be implemented as to have a dedicated thread to handle all incomming RX data, before it was done almost all in interrupt context, therefore the USB or transport bus usage is more responsible now.
