WebRTC: Use of multiple STUN servers in iceServers property of RTCPeerConnection for fallback

The question concerns the behavior of fallback ice servers in the WebRTC context. Here is an example:

  myPeerConnection = new RTCPeerConnection({
           iceServers: [
                          {
                             urls: ["stun: stunserver1.example.org",
                                    "stun: stunserver2.example.org"
                                   ],
                          }
                       ],

   });

The question I have is: For getting its own server reflexive address, when will the STUN client or ICE agent (as the case may be) switch from the first server to the alternate one? After the first STUN binding request times out or after all attempts, as provided in RFC 5389 Section 7.2.1, are exhausted? Or does it do a round-robin or some such strategy? I am, of course, talking about UDP transport and strictly in the context of WebRTC. There does not seem to be any specific clarification in any of the standards documents so I will be content with the current implementation/best-practices in browsers.