Anecdotally, the iAP listen for heartbeats from one another and when one (or more) is missed from the VC, the remaining iAP hold an election and a new VC rises.
My users have never noticed any outage or failure during such an election, and in lab we tried to create a loss of service, and were only able to when we simultaneously shut off several iAP. The outage was most likely due to the client having to search for a new AP rather than due to VC migration.