It's not about VR vendors forcing that onto anyone (that would silly and self-defeating), but rather developers doing it themselves to, say, eke out that little bit more performance.
The thing about that is that it would require multiple, very unlikely circumstances to happen all at once (making it even more unlikely):
1) You'd need developer with clear knowledge of the tracking-overhead & insight that they can *meaningfully outperform the SDK.
*
Substantial differences - not 1 or 2 fps more - noone would even consider what follows for that.
I've spent a fair amount of time with early evolution of modern VR, with launch titles across multiple platforms, and I'd be hesitant to say I could answer that question positively. Even though especially on PCs of the time (2016) - the SDK tracking overhead was really substantial (easily in the realm of 10% of the frametime available). We never saw that problem on console though, thanks to async-compute and SDK not being behind API walls that made a lot of things difficult/hacky in those days on PC. Not saying it's free - but it never came up on profiling as anything worth worrying about.
2) You'd need said developer to have someone on staff with significant experience working with sensor fusion and image processing/analysis - because we're literally talking about rewriting the tracking stack.
This isn't unheard of - but it's nonetheless rare enough that you remove 90% of developers out there by default to even attempt this.
3) You'd need to find a way to bypass TRC violations 2) would inevitably trigger.
I've been on that side of the fence, and yes there are stories I could tell about certain things being 'possible', but nothing even remotely on the scale of what we're talking about. We're talking substantial risk and I have no idea if it can even be pulled off (nevermind TRCs, just integrating with the rest of stack without everything falling apart would be a minor miracle). Moreover, getting caught would mean throwing out all the work from 1) and 2).
4) You'd need to convince your management that investing in 1-3 (we're talking multiple
man-years of investment here) is both somehow more worthwhile than anything else you could spend that $/time on, the risk of it all being thrown out in 3) is acceptable, and figure out a way to work that into your project schedule somehow without multiplying your development time.
Good luck with the above given VR budgets have been incredibly hamstrung for vast majority of projects, including 'AAA' ones (especially 'AAA' ones actually).