Well, the thing is: When replaying samples on only one channel, you get 16 volume values that are not linear at all, it's really an exponential curve, with most of the lower ones being very close.
So technically, when you go from 4 bits encoding (like what I did in Oric Tech) to 2 bits encoding, you really half the storage size, but you less than half the quality, which means that if you double the replay frequency you end up with a much better sound.
In this particular case, it was probably worth it, because 100% of the CPU time is being used to replay, but it's definitely something I need to investigate.
On a side note, the lsl/adc #0 trick is neat, but I think it would have made more sense to just encode it the other way so you can just shift right to get the next value