Multi-call memory contextualization benchmark