Alright, here's what we're going to try.
First off, we're not giving hard guarantees as to the maximum length of strings returned. Such a guarantee could create a situation where we might have to release an API revision just to accommodate increased text limits on the trilogy sites, or one - which is equally bad - where the API would return truncated results until the next "natural" revision.
Furthermore, spelling out exact character limits brings us into the realm of Unicode dragons; which, frankly, is asking alot of developers. This complexity can be seen full force in Twitter's API. People do like to use lots of Unicode, so it'd be a more common problem than you might think.
What we are doing is putting our current limits as
suggested_buffer_size in the documentation. These are, at best, hints. Any code you write must be able to handle larger data if its returned. This code will get pushed into production later tonight, in all likelihood.