http://search.cpan.org/~toddr/Locale-Maketext-1.19/lib/Local... - which is an article, not a CPAN module, despite its hosting. Localization isn't "find and replace" easy, it's a nightmare.

Not all languages will be as simple as s/his/their/ and s/her/their/.

I read somewhere that other languages had more neutral gender splits, or simply all-masculine ones, to make things easier for now. So it could be in other languages they had nothing to reverse yet.

But in English it is. It's acceptable to develop faster for certain (popular) languages.

Absolutely, but I would be surprised if they didn't have some code in the production version right now to support doubtlessly upcoming translated versions - perhaps not a full blown declension engine, but it seems unlikely they'd go English only at the start and retrofit everything later. The code might not be optimized to just let them do a simple find/replace for English.

The problem of translation should not affect the writing of the original text.

For all you know, the "original text" is a mix of "foo bar" and lolspeak.