The trouble with caniuse is that it only covers the "mainstream".
It does not cover more exotic things like browser in RiscOS or Amiga or text-only browsers etc. These are often consumers of text-only sites, deliberably or not. Small market share though I admit, but probably larger market share for text-only sites Vs other "normal" sites.
I think covering the "mainstream" is the point of the site. Don't get me wrong, I really do enjoy fast loading text based websites, but if web devs had to worry about those browsers, then CSS (and web technology in general) could never develop past its infancy. Whether that's a good thing or bad thing is up to you.
What I like about it is that the urls are "honest" -- you're not navigating to a new page (there isn't one, just like there isn't one when you do the same w/ a Jamstack App) -- and the urls show this.
It'll treated as one big long page with helpers to jump to each section. It'd be better if each section had it's own role (probably "contentinfo") and aria-label attribute. It'll probably work better than a lot of other 'modern' approaches even as it is now though.
Cool trick! If you search for random phrases in the pages that aren't the front page, Google has indexed them as well even though they're hidden by default by the CSS. I'd be cautious for SEO reasons as it's non-standard though.
> Firefox supports pure css masonry grid behind a flag.
That has less support than Bieber at a black metal festival though.
Here's the columns (or flexbox) way of doing it. Note that images will be a bit strangely ordered (as you would read columns). That's not solvable without JS (AFAIK). That's why W3C/CSSWG is looking at possible solutions that resulted in the Firefox experiment you're referencing.
I am tired of that too, but you can use the checkbox hack to implement features such as tabbed content for which storing user selections in the browser's history would probably result in unintuitive UX.