Sometimes I see an answer in the form David Roberts mentioned, the entirety of which is "See theorem 6.3 on page 357 of " linking to an article paywall or to a vendor of a book, and it will be flagged "not an answer". Actually it could very well be an answer, but it could be hard to tell if left in that form.
I think all users should be encouraged to provide content which can read and evaluated on the spot (and leaving a comment to this effect is how I typically respond to the flag). I would assume that whoever writes something like the above has access to said article or book and could therefore also write out the statement of the theorem, for the public good. Of course, MO is found so useful because a really good answer typically goes further, explaining the inner insight or what is really going on, but as far as baseline etiquette goes, I think it's reasonable to expect a poster to give some description of cited material that is not readable on the spot. (By "readable on the spot", I mean not just for professors who have institutional access to the major journals, but also for researchers who are more isolated and not close to a good university library.)
I agree that it's acceptable if an answer simply links to content elsewhere on the WWW that is readable on the spot. As I say, it would be even better if the poster adds some personal insights.
(Actually, this answer should perhaps go further to clarify "readable on the spot". For example, Wikipedia is banned in certain countries. I don't know enough about such bans to comment too much further, but I understand that some researchers have found the nLab very helpful because unlike Wikipedia, it's not banned. Google Books is another edge case because not everyone has access to the same pages.)