I tried to fix it by adding a 5G router for failover. That helped a lot, but the occasional drops still happen, and the resync cost is still there.
So for now, I’ve decided to keep all Longhorn replicas inside my homelab and just let the edge nodes consume the data over the network from home, instead of maintaining full remote replicas.
I’ll stick with that approach… at least until I come up with a better solution.