(0) We are ramping up the infrastructure soon to multiple servers. We've been running it on a single (but pretty high-end) server. We're moving to 4 virtual servers for fault-tolerance and load balancing. However, if (and before) this became official on MO, I'd feel better asking William Stein (owner of the servers, and the guy who started Sage) if it's all right to use his computers (if that's what you want to do).
(1) You could always set up a server that proxies requests to the cell server. For example, you could set up HAProxy to transparently reverse proxy requests, and at MO just use something like sagecell.mathoverflow.net. If our server goes down, you could redirect to a different server. Or you could set up your own server, of course. You'll also have the sage code, so you could set up something so that if for some reason the sage cell server isn't available, the page could still render the code as plain text.
(2) On the sage cell server, a user could do something like this and thereby embed whatever javascript they wanted into the page. You could prevent that by sanitizing the html that comes back from the Sage server, of course.
]]>(2) is indeed a big worry. Peter's reasons are reasons to worry less, but sticking user content in script tags is so terrifying you have to have more than a list of reasons not to worry. :-)
[1] (At the barest minimum, about $30/year, for an anemic processor that would be worthless to us. I'd guess for a few hundred a year we could run something usable, but I don't have a good way to estimate how much CPU a Sage Cell server really warrants.)
]]>re (1) Sage has been NSF supported for a while and is managed by an independent foundation.
re (2) Two reasons not to worry: a) Sage works server side so any code gets sent to Sage as pure text and is interpreted there and sent back. b) the text should be sanitized into pure text coming from the MO input anyway (not sure if sage would have problems with that)
]]>Much like MathJax, Sage really only needs a javascript call, so no major changes to the SE software should be needed (though probably a few minor ones to integrate the output). In fact, you could probably write a MathJax extension to wrap it all in one.
]]>