How does Multicast RPC handle actor relevance?

I found that actor replication is intelligently adjusted according to relevance.

This seems to save network bandwidth by adjusting replication between clients that are not seen as a problem with distance or location.

I want to know if the same replication throttling algorithm also applies to multicast RPC. For example, if you run a large, dedicated server with 100 connections, does multicast RPC called from the server request remote function calls to all clients?

The RPC won’t fire on clients for whom the actor that fired it is not relevant. So, if it’s something that sets a state that should be visible the next time that actor becomes relevant to those clients again, consider using replicated variables instead.
This post, from Alexander, is a good and brief explanation: OnRep vs multicast - #9 by TheJamsh - C++ Programming - Unreal Engine Forums