A recent patch (https://gerrit.wikimedia.org/r/327779) proposed to fix handling of the null character when present in language-converted text, to make it consistent with how null characters are handed when language converter is disabled.
@tstarling suggested a better solution would be to strip null characters entirely, whether language converter is enabled or disabled.
Indeed, the HTML5 spec frowns on null characters in HTML documents -- they are generally ignored or replaced with U+FFFD, and representing them via character entities is explicitly forbidden. It seems like good practice for the parser not to emit U+0000 in its generated output.