If they're having troubles with correct romaji transliteration (and even if not, to be honest), I don't see why they don't just transliterate it in furigana the

If they're having troubles with correct romaji transliteration (and even if not, to be honest), I don't see why they don't just transliterate it in furigana then. The source data of Tokeniser dictionaries such as NAIST JDIC isn't even in romaji, so if they are using a proper tokeniser, then they're actually doing an extra step and throwing away data to transform it into romaji form.