Agree with debjit_g, implementation is key. No doubt some optimized usb can be superior to less than optimized I2S. I've tried a wide variety of usb optimizations over the years, some better than others, same with dac implementation of usb. I'd suggest people look at the sophistication of usb implementation in my latest dac, Musetec MH-DA006, this not some cheapo chifi usb implementation, prior Musetec 005 used lesser Amamero implementation, Italians, 006 improves on this.
It is true external I2S clocking is not theoretical best, clock is best closest to I2S pathway within dac. Still, external clock superiority could be such that downsides of clock distance are overcome. In any case max I2S cable length should be .5M, less would be better. I agree implementation of any clocking scheme is important, therefore, not all OXCO clocks or any clocks are equal. If going I2S route one should investigate quality of clock in dac vs clock in ddc or streamer.
By the way, there are some streamer manufacturers coming out with I2S outputs.
As to what output/input scheme is superior, I'll maintain I2S has some inherent advantages. Therefore, how can it be an advantage for the dac to have to extract data from clock, the same with conversion from usb or whatever to I2S native path.
Bottom line, all is conjecture until one has actually implemented these optimized streaming chains into their system. My present I2S setup was a great improvement over my optimized over many years usb setup, this with three dacs, many others report the same. A prior less than optimal I2S setup I had many years ago was NOT an improvement over usb.