I guess if I set the cast shadows all the parts on the client then only the client will have to focus on doing the cast shadowing and the server can focus its memory on replication and physics handling more efficiently
The client handles rendering shadows anyways. As far as I’m aware, it won’t help whatsoever other than just giving you more things to do yourself that there are already implementations for.
The server does not do any rendering. The clients will do all the shadow mapping individually so you won’t have to worry about this. Even if this would work, don’t forget the golden rule: pre-optimization is the root of all evil!