Just pots being pots. This is typical IME, with the effect being determined by the pot, taper, and circuit. I think if you were to test this with many various pots on many circuits you'd find similar nonlinearities, especially at the extremes.
I don't think it's unintuitive. Just to follow up...