(At worst, a code-point in UTF-8 takes three bytes, compared to four in UTF-16 or UTF-32.)