is that an issue for the Threadiverse? I thought all replies are collected by the host instance of the community.

You're not wrong about that. Threadiverse software (Lemmy, Piefed, and NodeBB too) use something called 1b12 to synchronize content between instances.

If you follow a community, you set up the synchronization from your server to that community.

This is similar, but about backfill, which is what happens if you follow a community mid-way and don't have any of that stuff from before.

Also you can't 1b12 synchronize with Mastodon, so that unfortunately is a dead-end, but you would hopefully be able to backfill from Mastodon.