I ran a lot of tests while I am refactoring the website and found out how it works.
If you can name a font that exists on the user system, and set it through a stylesheet, it will use that and display hanzi or kanji appropriately depending on the font, and will IGNORE the language attribute.
If the font is not found, then depending on which browser you use, it may or may not select the correct font IF a language attribute is present.
The latter is called "Automatic font assignment".
More reading for those with insomnia :
Examples of language dependent characters:
http://en.wikipedia.org/wiki/Han_unifica...characters
Document-level Language Declaration, Test Results
http://www.w3.org/International/tests/re...eclaration
Automatic font assignment for CJK text
http://www.w3.org/International/tests/sec-cjk-fonts
This is already fixed in the "refactored" flashcard review mode. Though which font exactly it will use on Linux I dont know. But basically if you see the "language dependent characters" correctly above it will work. If not, then you need to tell your lazy browser what fonts to use explicitly, somewhere in Preferences.
I was amazed to discover, for example, that Opera doesn't support the Automatic font assignment, so the 5 columns of characters in the Wikipedia page above look the same on my system. Even Safari 3 beta doesn't handle it properly. IE6 and Firefox do.