Probably a bit of both.Before launch, SpaceX has the go/no-go poll long before webcast starts (at the start of prop load, over an hour prior if I recall right) so that means the count is already quite close to launch when stream starts. Beyond that, two fairly recent examples would point towards public countdown net not being used for a lot of communication:When F9 1.0 had an engine-out during ascent, it was apparently completely ignored on the countdown net. I noticed something was a bit off during the webcast when the timelines didn't match with the expected (stage separation later than estimated). I can't imagine everyone just being quiet over such an event...When Atlas V ended up second or two away from dropping Cygnus into the drink (mixture problem leading to early staging and Centaur having to make that up, which it did ever so barely - thank god for those margins), the only effect was that the callouts from NASA about upcoming Centaur shutdown were off by a lot, which made it obvious that something was up. Only later tweets explained the reason.I have no idea if this is due to the few second delay in the webcast (which gives them the ability to "mute out" stuff they don't want sent out to everyone or even outright cut the webcast) or because a lot of the communications is not broadcast - probably the latter. Maybe someone who actually knows can chime in and tell how wrong I am
Could part of it also be SpaceX using their Autonomous Flight Safety System? Fewer bags of mostly water on microphones.