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.
The feature is probably targeted at absolute beginners who don't even know kana yet. The screenshot of the settings does show the option to select a different transliteration, but romaji seem to be the default.
