Seems like the more readable the code gets, the worse it performs.

It's probably because of all the references and function calls.

lol. I've checked on godbolt and we've been getting less and less efficient (at least in terms of assembly length). My initial solution was 186 lines, then 247, and yours is 504. It's not an actual speed comparison, but assuming it's a similar effect, that would show an inverse relation to legibility.

Yours is definitely the most legible, whereas my initial minimizes the number of passes.