Well, on the server you're still starting N requests based on my batch request. I guess combined with caching that can make sense, yes. Even so, I'd probably want to implement my own [1] when setting up my own web app for reliability/trust reasons, and because of same-origin restrictions on client-side scripting. (XHR does support not following redirects, but it's kind of pointless with same-origin policy)
Yeah, it's only possible on the same hostname as the page you're on, which limits the usefulness and makes the server-side lookup necessary. The JSON solution will only work via a foreign host with a <script> tag as far as I know, which again is something of a trust issue.
[1] I did exactly that for http://tweetoftheyear.com/ in clojure: