Very good points. I think another is that we can develop theories that relate different experimental cases, so that a "replication" doesn't have to be precisely the same study, but different studies in a sense replicate different but overlapping views of the theory.