But recently I thought, "Well, it's surely all detuned, but those freqs are high and it's nearly impossible with a "timer" system to stay in tune in high freqs. Are those speeds the closest possible to tuned note frequencies?"
I took my Voyage200 and divided the 2A03 master clock (1.7897725MHz) by all the frequencies of the tuned notes, rounded it to the closest integer (so I get a value of what I'm supposed to put in the internal DPCM period register to be at the closest possible frequency) and I compared the result with the period of each speed.
The error is within -1 to 1, half of the time being 0.
Bananmos were not sure if "Nintendo designed certain values to be notes, or if it's just blind luck that they are". Well, there's strong evidence here that those periods were chosen according to how close it gets to some note's frequencies.
P.S.: If you really want to see all my calculations, I can post them on demand.