Ah, true. I wonder if it wouldn't be best if, by default, browsers would only complain when a site switched to an untrusted cert it hadn't used before (i.e. if Facebook.com is suddenly self-signed, that's a problem, and if a site switches from one self-signed cert to another, that's a problem as well). This would prevent most MITMs while still making self-signed certs useable. A site with a self-signed cert is better than one using no security at all, so it shouldn't get more flack.
[1] http://www.xmlgrrl.com/blog/2010/07/06/tofu-online-trust-and... [2] http://convergence.io/