The most natural time to have done this would've probably been going from XP → Vista/7, which broke a number of old things anyway.
For all that switching mainline Windows over to the NT kernel in 2k/XP improved Windows, the benefit wasn't as great as it could've been if they'd been able to unshackle the OS from backwards compatibility with all the preceding Windows/DOS stuff.
Microsoft had a good thing going, with the massive success of MS-DOS and Windows 95 it was imperative that the next generation consumer OS, Windows 2000, could run programs from their existing well-respected platform without users noticing the gap. (With OS X, you noticed the gap -- totally different UI style when in Classic Environment, and the Carbon API for easy porting was a mess that went unadopted by developers)
Really, they could have done a lot better-- a VM jukebox with everything from DOS 3.3 onwards available could have made the "most compatible Windows ever"-- even old software that assumed a 4.77MHz CPU clock could have been rigged to run natively.
The obvious solution is to leave those enterprise customers behind on the legacy edition of Windows and launching something new. But Microsoft seems incapable of going to market with something new also. They created Windows 10x, and then went and merged the new parts back into the legacy platform.