When you prune, you’re removing archived contracts from data stores on the participant and domain - does that improve ledger read/write performance?
The most important savings are in terms of storage and predictability with regards to their requirements. For the most part, usage of backing stores is optimized for the relevant services exposed by Canton regardless of the pruning state, so while you might experience improvements in latency and throughput with regards to the persistence layer by pruning archived contracts, that’s a secondary benefit.