However, if you assume all sequences are length 3, you still get 64 combinations.
We only use 20 out of that space. And if you look at how base pairs encode to amino acids, for half of them, only the first two base pairs even matter - since it's prefix-free you can guess the amino acid if you see those two and even ignore the third.
Given how underutilized this space is, I'm not convinced that increasing the domain to 216 will lead to much more than the ability to express our current amino acid space with only two base pairs.