I don't think that walled garden is technically applicable to the wild of the internet. Apple dictates (negotiates) the terms with the iOS carriers and can ensure that direct connections between iOS devices over the carriers is possible for the apps Apple wishes to whitelist.
Facetime over cellular wasn't enabled as a feature until last year, which means during the time period discussed in the patent lawsuit (last year) it was either non-existent or very new with minimal usage. Some carriers still don't allow it at all. For the most part it is/was a WiFi application meaning Apple has no special relationship. I'm pretty sure other applications will show similar statistics.
"Direct connections" can also be deceiving. This could include direct connections from one client to another, where that other client is acting as a P2P relay. Skype used this, because not all clients could punch a hole across NAT. So then you had some clients who are not reciprocating so that is the Tor model where some clients are relays and others are not.
I've not seen anything to suggest that Facetime can use significant bandwidth even when you aren't the one using it, and I'm pretty sure it would be widely reported since it matters to users with bandwidth caps and metering (this is of course well known for Skype). However, it is worth keeping in mind that when they say 90-95% of the connections are direct, that only requires that at least one of the two be able to accept incoming connections (or packets).
So the number of accessible end points is around 70%, at least for Facetime users last year (not necessarily representative of the entire internet). This includes both successful NAT traversal and no NAT.
BTW, no major crypto coins use NAT traversal techniques other than UPNP (sometimes). They just rely on one end or the other being reachable by an IP address. I2P support in Monero will likely improve P2P connectivity somewhat.