And what % of users are actually advertising their domain, running BGP, and operating their PBX? Now what % of users go through Gmail, Lumen, or AT&T and what happens when one of those go down? Which of these don't have centralized administration/regulation bodies? Which of these will a user say they haven't had hours of downtime with in the last year?

Even in the case you have a very well federated system with users very dispersed and the federation had 100% uptime and no bugs... how does it affect uptime for users? A central server going down 1% of the time for all users is the same as 1,000,000 decentralized servers being down 1% of the time for their 1/1,000,000 users. Nobody cares others can use the service during an outage they care they can use the service.

Federation capability at the protocol (even if it's not used in a truly federated way) provides a lot of great features, uptime for users isn't really one though. The most relevant is probably when Signal shuts down intentionally (tomorrow or 1,000 years from now, doesn't matter) you can't just migrate to a different Signal server without all of your contacts moving to the same one. This is akin to taking your /24 with you to your new ISP, porting your phone number to a new phone provider, or taking your custom domain email from Gmail to a different host and the ability to do that on those federated systems is probably why they remain in use today.

What exactly are you trying to argue here? "Federated systems fail as well, so we might stick with centralized ones for efficiency". Is that it?

If so, may I suggest you read Antifragile?

To point out that "the main lesson I wish developers (and especially Moxie) get to learn is the one that teaches them that centralized services are fragile, bound to fail and simply not worth the cost." has nothing to do with centralized or not. It's not a reason to do things one way or another. There has never been a large long running system, centralized or not, 500 page book or not, that hasn't had user downtime or didn't incur cost. This part is not a matter of theory nor what decentralized systems set out to solve.

There are other valid lessons for why choose decentralized but this is not a "ha, got'em" moment for that. There are also valid pros for centralized as well, hence their real world success and the reason most "decentralized" tech ends up being run in a very centralized fashion when it matures.