That's a really good point. However, it is easier to create the protocol in a homogenous environment than to backfill a heterogeneous one. Case in point: FaceTime. I wouldn't buy an iPhone if it could only talk to iPhones, because the precedent has been set before. I don't have a problem with FaceTime being Apple-only, because 1) there is no precedent and 2) Apple can potentially make FaceTime better by limiting the platforms they worry about.
There may come a time where a competitor has all the features of FaceTime (especially its ease of use) and have platform agnosticism. Then the decision might change.
If you go by "skate to where the puck is going to be", the puck, when it comes to communication, always moves towards platform agnostic protocols. (.. and Skype provided a decent fraction of Facetime's functionality well before Facetime).
If the likelihood of a person using platform A is P(A), then the likelihood of two of them communicating is of the order of P(A)^2. So even if you have 20% platform penetration, that only gives you a communication likelihood of 4%, leaving plenty of space for other providers to fill.
There may come a time where a competitor has all the features of FaceTime (especially its ease of use) and have platform agnosticism. Then the decision might change.