And the Gmail engineers should add an opt-out "high security" mode that checks the referer to make sure the form submission is coming from Gmail itself and not some outside website. This way people who like to use custom/blank referers can ignore this security concern if they want, and all the rest of us can prevent the risk of this problem.
And then when sending AJAX or form requests in the background, make sure to include that as a request parameter. That way, even if the user has the right authentication cookies, external websites won't be able to fool Gmail into thinking they are Gmail.
Really, this doesn't seem like a very hard problem to solve...couple lines of code...
They store information per logged in user but are they performing some type of lookup information for every single AJAX request? I opened up a chat box and 7 HTTP requests fired off. I have no clue what these are for, but are each one of those validated and could they be quickly? During typical gmail usage several requests are fired off in the background for simple and seemingly trivial user actions like the one mentioned above.
EDIT: ...and after I was done typing the message above I returned to gmail and 15+ more requests has been fired off without me doing anything. The message took well under a minute to type. Seems like a lot of validation would have to be done.
EDIT: To answer your question of "are they doing some kind of lookup for each AJAX request?" Well of course, since they already have to look up a user's ID, account information, etc. based on the cookie they send.
Agreed they have to authenticate but I don't think they perform a lookup on every single request. They probably use a key but it is more likely this key (let's call it the AJAX key) is generated from the user's dat (id, ip, whatever else) using some hashing function on the back-end. When the server receives a request it can check to see if a request is valid by re-generating the AJAX key from the requests meta data (header or whatever other data sent with the request for authentication purposes). This is much quicker and more scalable than a lookup for every request and its just as secure. Even if someone guesses your key generation method (which should be HIGHLY unlikely) you can simply change it and it will work for all users immediately, even ones who are logged in already.
Now maybe they are performing a lookup for each request but I just do not see how they can handle the load, or why they would want to given the alternative I just suggested. Google sends a ridiculous amount of data back to their servers during a Gmail session. I haven't examined the traffic extensively but open up a chat box in gmail and the Firebug console at the same time. Click anywhere in the chat box. See the request that fires off int he console? That happens EACH TIME you click ANYWHERE in the box. I guess they are doing some type of clicking heat map or something I don't know, but whatever they are doing it requires sending potentially 1+ AJAX requests per second for many users.
I am not knowledgeable enough to tell you how much is too much for a server/file system/database to handle quickly (~150 ms per request for millions of requests at once) so maybe the situation I just described is not as bad as I made it seem.