TFA is really talking about two things. I responded to the "layered on top of HTTP" point, while you seem to be talking about the "silos of end user data" point. Perhaps the first point contributes to the second, but I'm not convinced. Users could choose http apps that don't silo.
It strongly tends to, because having your own computer (especially a phone) set up to answer inbound HTTP is both administratively difficult (fiddling with NAT and dyndns, etc) and a security risk.
This is Schneier's argument about "security feudalism". And to some extent touched on by the article - people build on top of well-understood transport layers in order to reduce risks, and novel protocols are firewalled in case of novel risks.
