Nonholonomic constrains: why does not the least action principle leads to equations describing the motion consistent with the physical behaviour?

The least action principle seems not to lead to equations describing the motion consistent with the physical behavior for nonholonomic constrains. Here an answer to this question in proposed.