To explain further in the context of WebRTC, traffic analysis of the encrypted data could determine the pace and duration patterns of speech between multiple speakers. This could:
(a) narrow down or uniquely identify each party
(b) provide information on the mood of each party (are they interrupting other speakers more often than usual?)
(c) guess the nature of the call (information dump from one person to another, one person quizzing another person, ...)
(d) determine the languages used in the conversation
(e) guess demographic information from the conversation including approximate level of education/intellect, age, male/female, ...
(f) narrow down the physical location of speakers who are attempting to mask their identity through intermediate nodes
The following papers are a good starting point:
[1] Guessing the URLs being browsed by users over an encrypted TLS session: https://research.microsoft.com/en-us/um/people/gdane/papers/...
[2] Guessing the co-ordinates of where a user is scrolling around on Google Maps over an encrypted TLS session: http://www.ioactive.com/pdfs/SSLTrafficAnalysisOnGoogleMaps....
[3] Guessing the content of encrypted VoIP conversations: http://www.cs.unc.edu/~amw/resources/hooktonfoniks.pdf
[4] Guessing communication paths on the Tor network with only a partial view of the network (not strictly related to encryption but the principles of traffic analysis are relevant): http://www.cl.cam.ac.uk/users/sjm217/papers/oakland05torta.p...
[5] Guessing passwords sent over the SSH protocol using keystroke timing analysis: http://users.ece.cmu.edu/~dawnsong/papers/ssh-timing.pdf