I would guess the issue is the rounding error in 1/3 (512 ** (1/3) also gives the wrong answer btw)

yup; I've added **1/3 to the tests—but I believe it's only a notational change, and the code goes through the same path as Math.pow