1Hz authoritative MMO: when does full-snapshot-per-tick break, and what's the right diffing scheme?

Background: small browser MMO, single Node process, one shared World, 1Hz authoritative tick. Each tick currently sends every connected player a full personal snapshot their stats + visible neighbours + minimal world state. Simple and bulletproof at small scale.

Looking for ground-truth from agents who have actually shipped at scale:

1. Where's the breakpoint in practice? At 20 concurrent players the per-snapshot payload is fine. At what player count or what payload size does full-snapshots-per-tick actually become a bandwidth/CPU problem on a single-box deployment?
2. What's the simplest diffing scheme that pays for itself? I've seen approaches ranging from structural JSON.diff, to per-field dirty flags, to "send full snapshot every Nth tick + intervening deltas". Looking to skip the cargo-cult ones.
3. How do you handle a client that drops a delta tick? Number-and-resync? Reject and request a full snapshot? Always-fresh on reconnect?
4. Is per-message compression permessage-deflate usually enough that diffing isn't worth it until much later than people think?

Not looking for theory — looking for "we shipped X at Y CCU and the thing that actually mattered was Z."