For now I am still not keen on webfonts, as my primary language is Chinese, which, even using simplified form, would need at least 1000 popular characters. Unless there is a even more dynamic way to load fonts by splitting font characters by usage frequency, it would be a no-go for that (who uses webfonts in the size of megabytes?)
There are tools which allow you to subset fonts, splitting them up by character, so you could, say, produce a font containing only those characters that you have in your headlines for you headline font. For example, Font Squirrel http://www.fontsquirrel.com/fontface/generator will allow you to chose particular characters to include in your subset; though that's not very useful unless you really only need a handful of characters that you can enter manually.
It's true that for practical reasons, web fonts are less useful for CJK characters than they are for languages with smaller character sets. I imagine someone will write tools to make it a little easier to deal with, but it may not ever be as convenient as for other writing systems.
Well I am aware that this is possible, but only useful if I can somehow "combine" subsets into one single font. I have not yet devoted time to see if this is actually possible right now, though.
Say I have the lists of usage frequency for my target language (say Simplified Chinese) and then I want the most used 60% in subset A and other 40% on subset B. But at the end I need to have 100% applied to the same font, which doesn't seem possible without some javascript mangling by looking up the text to apply the correct subset ...
