for 15 years now, I have been asking students to use datasets large enough to require efficiency.
like, auto-complete with the full english dictionary, search on real point clouds with millions of points, or TSP on city lists of real countries.
e.g., if they do not use a hash table correctly (implementing comparison functions, not just == ), and perform dumb linear search, it will take 100s of seconds. if they implement stuff in a stupid cubic way, it won't finish in the same month.
and something I really found amazing:
conspiracy theory "USA did not go to the moon" because "there was no way such old computers could control a spaceship - they could not even run a VGA screen!"
just lack of understanding that once the equations are solved in the paper, the actual control laws aren't really computationally intensive (at least for today's standards), a few kflops could handle that.