My point was not that a commenting system should be implemented using this or that protocol (though it's likely just as easy using established AP or web<->XMPP frameworks). In fact, i believe a public commenting system is a very valid usecase of matrix's censorship-resilient decentralized rooms even though i'm not entirely sure of the consequences of that in terms of moderation/spam.
I was reacting to matrix being "not meant to be any one thing". I explicitly recall matrix being marketed by the community (maybe not the devs themselves) as a modern, censorship-resilient IRC replacement that fitted in a short (single?) specification and intentionally avoided the extensibility (and associated implementation/interop failures) of the XMPP protocol.
When i say matrix is a more specific use-case than other federated protocols, i mean that decentralized rooms can be implemented as a consensus-reaching algorithm on top of any federated protocol, and that's in fact what matrix servers are doing under the hood. But supporting the usecase of least-metadata-leakage in a protocol designed for sharing state across many actors is arguably trickier.
For example, i believe matrix doesn't currently support per-room nicknames which don't reveal your public address to all members of the room (only to chatroom admins for ban purposes). matrix has very interesting developments with or without this specific feature, but i was highlighting that matrix is not more generic/agnostic than other federation protocols (just like XMPP isn't a "universal" protocol either).
Like i'm very interested in matrix P2P ecosystem there's some really amazing stuff being developed there (pinecone), but i must say the entire matrix selling pitch is very similar to the selling pitch of XMPP more than a decade ago: "a universal bridgeable messenger". Regarding the P2P example, XMPP had offline-first "zeroconf" federation (XEP-0174) drafted in 2006. Despite being far less advanced than modern matrix P2P, it was already very similar in spirit.
So my central point i guess, is not that one protocol is better than the other. They all have very strong pros and cons depending on the actual usecase. Different users, or same users across different contexts/activities may prefer one technology or the other. My point is i believe it is our responsibility as technologists to ease their life and standardize things for more interoperability so users can have a choice between "the federated networks" and "centralized silos" instead of having a choice between "centralized silos" and "tiny federated islands that mostly don't talk to one another", adjusting the balance of power in our favor which is in the direct interest of everyone involved except the corporate silicon valley sociopaths.
Cory Doctorow's latest talks have pretty compelling arguments for interoperability if you have some time to spare.