The theory is that the biggest source of error in riding calibration courses is the inevitable "wobble" at the beginning of the ride. It's very difficult to ride perfectly straight when you are first start from a standstill. With a very short calibration course this wobble error will make a big difference in the overall calibration count. For longer calibration courses it will make less difference.
When I measured a track with a measuring wheel, I calibrated the wheel with a 100-meter calibration course, because I reasoned there was no "wobble" at the startup. A 20-meter calibration course probably would have worked just as well.
Over the years several people have questioned the 300-meter required length of calibration courses, pointing out the fact as you have, that there are other sources of error that will be bigger for long calibration courses than for short ones. Having to use multiple tape lengths is one, however I tend to think that the small allowable tolerance between the two measurements of the course helps with that one. For me, the more concerning error source is the fact that in some parts of the country it is difficult to find a good location for a 300m calibration course. So people end up locating their course in a location with a slight hill, or where the road isn't perfectly straight, or where the surface isn't typical (concrete instead of asphalt). If they could set up a 100m or 200m cal course they could probably find a better location.
Is the error caused by the above issues bigger or smaller than that caused by startup "wobble?" That's the question.
Here are some threads from years past addressing this issue.
https://measure.infopop.cc/top...-calibration-courses
https://measure.infopop.cc/top...s-unnecessarily-long
https://measure.infopop.cc/top...ation-over-25-meters
I also had a thread reporting results of an experiment I did with 3 different cal courses (100m, 200m, 300m) in the same location. I did multiple rides of each one and compared the cal constants. Unfortunately, I can't find that thread.