Assignments are either for helping the students learn, or for evaluating them. Grades are either to give them an incentive to do the work involved in learning, or to evaluate them. Two very different goals!
For evaluation purposes, grades give the maximum information is the distribution has maximum entropy, so ideally there ought to be about 1/3 A's, 1/3 B's, and 1/3 C's.
For evaluation purposes, assignments are most useful if they help distinguish students from one another, so the ideal exercise is at the right level of difficulty so as to be solvable by exactly half of the students. An exercise that is successfully done by all serves no purpose for evaluation. An exercise that almost no one can do may still serve a purpose: that of distinguishing the few students who stand out by their ability to solve a problem that the vast majority could not do. Exercises with a variety of difficulty levels help evaluate more finely. Since the primary goal of midterms and finals is to evaluate, that's how they should be designed, and the median grade for those would probably be around 50. In fact, that's more or less how things are done in France. The learning part happens in exercises that are not graded, and the sole purpose of grading is to evaluate.
In the US, my experience is that assignments are usually designed so that, with enough time, most students can do most of the assignment: their primary purpose is to help the student learn. A very different perspective!