I wonder if Microsoft has some kind of internal metrics that show that every time they increase the major version number for Windows, a statistically significant percentage of software falls over because it performs its own (broken) compatibility testing. Limiting the number of major version changes and primarily using minor versions to distinguish between retail releases would be the path of least resistance for them and for third-party developers.

Edit: I read Raymond's old blog post there, and it looks like it was definitely a problem for Windows 95 so it wouldn't surprise me that it continues to be an issue.