As I see "ICE candidates" are successfully found on both endpoints (I look at chrome://webrtc-internals), the media sources are set up, but then some magic doesn't happen. It seems irrelevant whether I request video calls, or audio-only calls.
chrome://webrtc-internals only shows a single peerconnection being created so something is very wrong. Looks like (from the websocket frames) that when calling 987 from 123 there is a proper jingle-message-initiation message sent. Then there is a response but the actual jingle session-initate sent by 123 never reaches 987.
I'd suggest testing with http://legastero.github.io/jingle-interop-demos/stanzaio/ (or the strophe version)
Also, you are only providing a stun server. ~20% of calls will not work without a TURN server. But you only need to worry about that later.
