Even if true, a bimodal distribution of suckers and winners is very different from having to meet a treshold in certain skills to be able to code at all. Not everybody can become an architect or a mechanical engineer either, but I never hear my engineering buddies talk about the mythical 10x civil engineer.
Those statistics on marks in a classroom are the first hard stats I've heard of indicating a bimodal distribution in coding skill. I have no idea where they come from though. Do you have any links to sources?
The only numbers I've seen come from the Separating Programming Sheep From Non-Programming Goats papers by Saeed Dehnadi and Richard Bornat:
> We have discovered a test which divides programming sheep from non-programming goats. This test predicts ability to program with very high accuracy before the subjects have ever seen a program or a programming language.
http://www.eis.mdx.ac.uk/research/PhDArea/saeed/
It's entertaining, but I think the double hump in first year comes more from Dunning Kruger and compulsory courses than anything else. I've heard of the same kind of distribution in first year biology classes. Even if "true", I don't think it gives us an excuse to anyone we think is insufficienty spectacular out of the industry.
Dehnadi and Bornat later walked back their claims.
http://retractionwatch.com/2014/07/18/the-camel-doesnt-have-...
Furthermore, there is evidence that the ability to code is non-linearly dependent on some fairly uniformly distributed basic abilities. The bimodal mark distribution of first year software courses is evidence of this. It is trivial to reproduce it using the assumption that the underlying skill-set is uniformly distributed, but there is a threshold of ability below which coding becomes very hard: http://www.tjradcliffe.com/?p=1471
There is a very simple way to disprove this hypothesis: find a way to teach coding that will get rid of the bimodal distribution of marks, which has been vigorously attacked by educators for decades without progress. The things we have learned in that time are:
1) It doesn't matter how passionate the teacher is about believing that "everyone can learn to code"
2) It doesn't matter what language is used
3) It is very hard to predict who will be good and who will be bad.
Because of this, there are three reasons we know cannot be the source of the problem: the attitude of the teachers, the langauge used, or the number of years experience prior to university. If any of those were the cause the extensive research on this issue would have revealed them in the 80's.
I'm prefectly willing to believe that most people can be taught to code, and if I can be shown evidence of a teaching method that gets rid of the bimodal distribution I'll happily abandon the barrier-to-entry model I've proposed in the link above, but until then, it can't be called anything less than a plausible hypothesis. Certainly not a "myth".