Talks are the occasion for me to wear hawaiian shirts and/or rats. Maybe this is why I enjoy them. And maybe people like my talks because of the french accent, and because I try not to forget to thank them very many at the end.
Fun apart, I think talks have at least the same value than papers when it comes to live research. You cannot say in a talk everything that you can write on a paper (and you shouldn't even try, if you want my opinion). But the reverse is true: there is a lot of Jedi tricks that you can use in a talk (and not in a paper) for making people understand what you are really doing, and sometimes, when you're lucky, you may convince one or two guys to have a look at your research, or even better, read one of your paper.
Sometimes, I give talks that are only vaguely related to a paper. Or not related at all. And I tend to enjoy those even more. This is why I propose you a list of some lost talks.
Super-scalable P2P Networks is a talk I gave at the 2012 LINCS workshop to tell about a joint work with François Baccelli and Illka Norros. Tight schedule to cope with and why-is-that-bloody-video-not-working syndrome, so you should fasten your seatbelts or wait for a longer, smoother version.
Ten Things I Hate About CCN is a 2010 presentation that details why Content Centric Networking may not be the Ultimate Answer to the Ultimate Question of Life, The Universe, and Everything. May contain traces of bad faith.
Collaboration and Incentive Strategies in P2P Networks, February 23th 2006. This was a keynote presentation for the ICIW 2006 Conference. Turned out to be yet another bogus conference, but I didn't know at the time. Hey, when you're offerred to give a keynote just a few months after your PhD, you don't look twice. Retrospectively, that may have been a hint. The talk, which I cannot find anywhere, was one of the first ones about Preference-Based Systems.