I am too, and it seems testable, at least for large companies. Are the salaries bimodal within companies or not? If not, then the hypothesis would seem to hold water.