When the email is sent, there is no record of the message. There is only an image URI that points to the image that contains the message. So, the NSA won't be able to grab a copy as it's sent. However, if an organization has the ability to intercept all images that are loaded in the browser (through the ISPs, for instance), it can grab a copy of the image containing the message. In that case, some OCR would have to be done in order to extract the message (which, agreed, is not hard to do if you are explicitly looking for messages contained within images).

+1 That was my exact thought.

If anything; I think this gives people an impression that they can NSA proof their email-- when in fact they cannot.

The NSA is capturing, and archiving, the data directly from AT&T and Verizon's network. The fact it's set to expire on the OPs servers is of little relevance on theirs (NSA).

Cryptography is the only way to "NSA proof" anything; and only if you do it right.

Exactly, they're allowed to store the email, even if ostensibly they're not allowed to read it. But if it ever becomes relevant, they'll simply go back to their copy of the email.