I'm also a fan of unicode and i18n, however it doesnt add quite as much to security as it may seem.
the whole unicode codespace is only 22.1 bits, meaning that a unicode character is only adding about ~3 ascii characters
worth of entropy to the passphrase. It might be easier to memorize a larger number of bits. But if you know the native
language of the person who owns the password, you can eliminate the vast majority of the unicode code space from the
search space, most likely resulting in only 1 or 2 bytes of entropy per character. (more for languages with a larger
number of characters) However, a dictionary search will bring the entropy back down to the same or less than ascii,
unless the user uses random non-dictionary, or complex phrases.
There are additionally difficulties:
Some languages are not very easy to input without being able to see what you are typing.
A strong security system will prevent shoulder-surfing by showing circles, or even better nothing at all, as you type
Trying to HenKan Japanese or Chinese without being able see what is being written can be challenging.
In addition, using a device or computer without a given input method would make it virtually impossible to login. Not
every machine is setup with your idealized input method, or trained to your writing style.
In short: the downsides seems to dominate, and there is no significant security advantage. I use unicode passwords myself where
they work, just because I like them. (especially for throw-away accounts) I do think that software should accept passphrases in
utf-8, just for completeness sake. But I don't really think that they improve anything...