Conversational recommender systems solicit feedback from users in order to objectively inform the recommendation process. Ideally the user is presented with suitable products/services as promptly, as possible. Efficiency is key, and normally, this is measured in terms of the session length (i.e., the number of recommendation cycles with the user). In this paper we argue that it is also important to understand the effort required of the user during these interactions. Cognitive load refers to the level of effort associated with thinking and reasoning. We will look at the cognitive load implications, as measured by interaction time, of a critiquing conversational recommender which uses dynamically generated compound critiques. In particular, we find two interesting results. First, on a cycle-by-cycle basis the dynamic critiquing approach places a greater cognitive cost burden than that for the unit critiquing approach. Secondly, and arguably more importantly, the reverse is true when we look at overall session performance – that is, the dynamic critiquing approach outperforms the unit critiquing variation. We demonstrate these in relation to results obtained in a recent real-user trial.