Minimize networking overhead - seriously. Use client ↔ server communication only when it actually matters, and send as little data as possible each time (that includes setting properties of an instance on server)
“Networking libraries” often get it wrong. Having multiple RemoteEvents is actually good - it reduces dispatch overhead since you’re not shoving arguments into one event for the server to sort through.
Pack data using bit32 or tight binary buffers - algorithmically, not with “serializer” libraries. Nothing is magic. If you treat something as magic, you’re already losing optimization you don’t understand yet.
Take parallel into consideration (especially on a server if it has so many people in it); honestly, parallel would be your highest priority to take for almost everything when you don’t do write-on instance operations.
For computation or buffer operations (or bit shifting), consider using native code generation.
Consider using micro-optimizations; even 0.1-10% adds up a bit when you have such a huge server.
If you’re serious about performance, learn bytecode - it’s the only way to see what’s actually optimized under the hood.
And yeah, avoid bloatware modules or “single script architectures.” Those patterns look “enterprise-y” but perform like a brick. Only trust what you can prove through profiling and benchmarks.
Oh, and yes - StreamingEnabled is a must, but that’s the easy part.
Final tip: move any repetitive read operations off serial execution to parallel if possible.
There is a post I made where I and other people had sent some optimizations you may need https://devforum.roblox.com/t/code-optimizations-sharing/3882214
Maid and Signal is literally the opposite of optimization; what are you talking about?