The odd naming is consistent with MS's history of Windows naming, too. Here's the consumer Windows product line since 1992, ignoring high-end options like NT and Chinese-only releases like Windows 3.2:
Windows 3.1, Windows 95, Windows 98, Windows Me, Windows XP, Windows Vista, Windows 7, Windows 8, Windows 8.1, Windows 10
I mean that sort of goes up as a general pattern, but only generally.
In the case of Windows there were two (maybe three) different product lines all called Windows. The naming makes slightly more sense in that context.
* Windows 1/2/3.0/3.1/3.11/WFW3.11 - 16-bit (with 32-bit extensions as win32s), ran on DOS
* Concurrently, NT 3.1, 3.5, and 3.51 were marketed for servers instead of home users. They were fully 32 bit and separate from DOS.
* Windows 95/97/98 started naming after the year. This line eliminated DOS (from the user's perspective) and was 32-bit without the need for installing extensions.
* Concurrent with Windows 95 was NT4 which had 95's new interface based on Cairo
* At this point it was clear NT was the future for the desktop, so keeping with the year naming came Windows 2000 (which was really NT 5.0, but desktop users were by now used to year naming). But there was one more release of "old" Windows, as Millennium Edition (ME)
* The two names that dont fit are XP and Vista; these are NT 5.1 and 6.0, respectively. And now Windows 7 and 8 make more sense; we are back to the versioning we had starting with NT 3.1
* To make things more confusing, the server versions now started getting named after the year when previously that was the desktops. So we've got Windows Server 2000, 2003, 2008, 2011, and so on.
* Windows 10 is the oddball, skipping 9. I like to think someone at MS was poking fun at Apple, which skipped OS 9 to go straight to OS X, which two years later got the version 10.0. Mac OS 9 came after OS X but before OS X 10.0. I recall reading that 8.5 was originally supposed to be end of the line for classic mac os, but I don't have a source. So now there's 8.5, X 1.0, 9.0, and X 10.0, and everyone seems to think the X is a 10.
Yeah, I'm aware of the history. It adds more context to the mess but doesn't really make it sensible.
As for why they skipped 9, I believe I did hear a good reason for that one: people sometimes tested for Windows 95/98 by seeing if the version started with a 9, and they didn't want to clash with that. So this weirdness is simply an unintended side effect of their inconsistency in naming/versioning scheme.
I've heard that too, but can that be true? Wouldn't a lot of stuff have stopped working on Windows ME (the next edition after Windows 98) if that were the case?
No, this wasn't for Windows executables working or not working in a core binary technical compatibility sense. This was for where either an app or a website needed to test the version number against several possible versions of Windows to determine which version of Windows they were running on and react (or reject) according to the programming.
For later versions of Windows, programmers applied other tests, but none of those tests stepped on the "does it start with 9" test for 95/98. Using 9 for the version after 8.1 would have done so.
Windows 3.1, Windows 95, Windows 98, Windows Me, Windows XP, Windows Vista, Windows 7, Windows 8, Windows 8.1, Windows 10
I mean that sort of goes up as a general pattern, but only generally.