Some queries might be too slow at p95 even on a read replica with no other clients. Those kinds of queries can benefit greatly from a materialized view, and incremental view maintenance as data changes.

Not denying your point but in some applications (which aren't outliers) a good cache can increase number of requests served 100x. Achieving those with read-replicas is easy and possible but can be quite expensive.

A read-replica doesn't give you perfect transactional guarantees as well. Read operations to the replica after a write operation to the leader might still give you stale data (lagging by a few milliseconds).

Yes to a degree, but not if you require/value consistency.