A likely explanation for this is that CHRTRAN() is modifying the string in-place. If it encounters a 'missing' substitution character then it is probably closing up the gap with a bulk move of the remainder of the string left one byte. Starting with 6-million-odd bytes, in this case. By the time it gets the end of the string it will be running quite fast (only a few bytes left to move).

It could be made faster quite easily, by getting it to shift only one byte over at a time. The question is whether one of the fox developers is willing to potentially break things trying, this late in the development cycle.