It sounds like you're arguing from a hypothetically perfect view with 20/20 hindsight. Scaling problems are good problems, solvable at the right time without the evils of premature optimization. Doing work without a need is wasted effort.
I'm just stating the obvious, which is that whether a service is centralized or federated has nothing to do with it. Mastodon would melt long before operating at Twitter scale, for example. (And to your points, I don't think that's necessarily bad.)
But, if there's any company on earth that should have 20/20 hindsight about this, it's Twitter. Post-"fail whale" era and pre-Musk, the service weathered far larger challenges without collapsing.
E.g. signups have hit ~6x what they were earlier today but it's still only an average of one signup per instance per hour.