However, it is the sort of thing that creative hackers are good at combining with other capabilities or exploits in order to get at data or code they shouldn't. The trend in web API design is to keep things same-origin by default, and give control to authors over whether they want to give other origins access to their data, because this keeps things secure by default and limits the number of tools that those creative hackers have at their disposal.
Not everyone at Mozilla agrees that this is the right solution. Anne van Kesteren (formerly of Opera, now a Mozilla employee), for example, favors adding a "From-Origin" request header that would remove the need for servers to add Access-Control-* headers in many cases (though it would not provide the same security properties by default):
http://annevankesteren.nl/2011/02/from-origin
http://www.w3.org/TR/2011/WD-from-origin-20110721/
Meanwhile, some Google employees have been major proponents of making things same-origin by default; read the Tab Atkin post linked above for example. This isn't really about sides; it's about constantly trying to find better ways to improve the web without compromising security. I won't claim the current solution is perfect; it's a starting point, that's all.
Thanks for the debate, I learned a lot. =)
Good point but I believe that the only way to combat that particular edge-case is to use something like the valid_referers directive: https://gist.github.com/jdorfman/6737413
"This isn't about a known attack, but a class of attack that we know from experience is likely to crop up in the future for any type of cross-origin resource load that we allow."
You bring up another good point, but I am not talking about every cross-origin resource here, I am talking about Content-Type: font/ttf, application/x-woff, application/vnd.ms-fontobject & image/svg+xml.