It works fine if,you set your main font to something including Greek, e.g. Brill. All you're missing in that case is correct hyphenation. Alternatively, you can set a font by Unicode range rather than language (it doesn't do this automatically).

It strikes me that what we really need is a filter that would tag languages in Pandoc output based on best guesses (Word does this to some extent already). Should theoretically be possible with the language detection libraries out there.