Taken in isolation, without any way to know critical thresholds, I would argue that neither is useful. If someone worked out the critical thresholds through experimentation, I’d probably go with the Kelvin number, if only because no thinking required. With the other approach, it might be fine, but I’d have to think harder to be sure. So, based on expediency and conservation of brainpower, I’d go Kelvin. That’s not a proof that Celsius would be wrong though. If you like puzzles, and have nothing else to do, see if you can prove it. I’m guessing you probably can.
[Edit: I think we can all agree that Kelvin would be objective truth. As a possible quick answer, I’d try to see whether I could find two pairs of celsius values with the same percentage difference in Celsius but a different percentage difference in Kelvin. If I could, then that would be proof by contradiction that the Celsius method isn’t as accurate. ]