He talks a bit about how to pick the right number of successive letters to use as hash keys - which is where you can get a handle on alphabet sizes. I would guess (maybe it actually says) that he Wikipedia dump in the benchmark was UTF-8 or ASCII and, either way, treated as an alphabet of 8-bit characters. The DNA case is kind of interesting (2 bits min but more likely 3 or 4 in a typical genome record).