First, the scenario you describe isn't workable as it is the professor who is ultimately responsible for the course. A problem that is likely to arise is that the quality of lectures would be spotty, with some being fine and others not. I'd guess that students would notice this and complain about the variability. It is also an imbalance of time and effort. Lecturing takes up relatively little time compared to grading. A professor personally grading 200 programs seems impossible if any feedback is to be given. And with five TAs or so, I don't see a way that work load can be balanced. Five lecturers every day? Probably not. TAs only working one day a week for an hour or so while the professor grades 200 papers every week? Unworkable.
A deeper problem in STEM fields is that graduate students may have an imperfect understanding of the topic (though professors might also, I suppose) and, through lack of deep insight mislead students. I developed very deep insight into real analysis (I'm trained in maths), but not really until I'd finished my doctoral research. Up until then I had a somewhat mechanical understanding. But it is that insight that lets a professor guide students toward a goal.
Related to this is that TAs, assuming that they are students, may have the misconception that lectures are about facts rather than about driving the learning of the students. It is too easy to make assumptions about students (and their understanding) if you are one yourself and are biased by your own learning modes and practices. New professors sometimes have the same problems, as I did. Experience, however, gives you a better and more productive view. I've given one absolutely "perfect" lecture from which the students learned exactly nothing. I learned my lesson. One hard lesson for professors to learn, though they must if they are to be effective, is that nearly every one of their students is not like themselves and learns differently than they did and has different goals than they do. TAs are unlikely to have this sophistication about teaching and leaning.
To answer your objections to the "standard" practice, note that it is the responsibility of the professor not only to lecture, but to provide a proper grading rubric that can be followed by those less adept in the subject and that can be justified to students. Lacking that, chaos can ensue. One way to develop a rubric is for the professor to look at a few papers randomly before making the rubric so as to see what sort of guidance the TAs need to follow and what sorts of messages need to be given to the students.
Another adaptation that reduces the variability of grades given by TAs is to have each TA responsible for some of the questions for all of the students, rather than all of the questions for some of the students. That is how we handled major exams back in the day (last century).
Another good practice is to have the professor meet with the TAs as a group and encourage them to bring up any difficult issues they have seen since the prior meeting. The professor and TAs become a team.
However, there is some value in letting (or requiring) TAs to conduct occasional lectures. If a professor gives a TA a couple of weeks to prepare a lecture on a specific topic, reviews their lecture notes, and attends the lecture then it can be beneficial to the training of the TA without being disruptive to the students. I was given such a task by a professor even as an ordinary student in a course. As an undergraduate it was a very difficult task, done fairly poorly, but a good experience. I got feedback from both peers and the professor.
