From John Baez at the n-Category Cafe: the Journal of Number Theory is now inviting authors to post their abstracts on YouTube.
Something about putting them on YouTube -- as opposed to on the journal's website -- strikes me as saying that they're not "really" part of the paper.
More interestingly, though, the few "video abstracts" there are just people reading their abstracts. I think it's a good idea, but mathematical speech is just not mathematical writing read out loud. (This is, of course, the converse of the fact that mathematical writing is not just mathematical speech transcribed.) Perhaps there's potential here, though. The first few minutes of a good talk would work as a good "video abstract", I think.
But you've got to start somewhere.