You're using a bucket size which is a power of 2. A random byte fits on the interval of [0, 255] which is evenly distributed modulo 256, which if you reduce mod 2^N for 1 < N < 8 will still produce a uniform distribution.
If you copied my JS code into your console and changed the 54 to 64 the bias evaporates. Hooray :D
However, for arbitrary ranges, you should use an unbiased selection strategy to ensure a uniform distribution.
http://3v4l.org/EgfPR - notice 0 is significantly more common than the other values when reduced mod 17
