Not sure that I entirely agree there: I could easily imagine someone writing too close to the metal to stick to list iteration where a hash would be better and

Not sure that I entirely agree there: I could easily imagine someone writing too close to the metal to stick to list iteration where a hash would be better and so on, unless it's prohibitively slower, whereas someone slightly higher would freely choose whatever they feel appropriate for the situation. They might often guess wrong and take a structure that is overkill for the typical dataset but the penalty for that will be negligible compared to the one paid for using a badly scaling structure on way too much data.