> You're gaining a great first impression at the cost of a poorer experience for every subsequent page load.
Yep, it's unfortunate we still have to make tradeoffs like this. At least in this case, it's less than 8KB added per page vs something more complicated that might break.
HTTP push was getting closer to offering some alternative like what you're suggesting (the server pushes the CSS file to the client in parallel to the initial HTML page, and the client can say if it already has the CSS file) but it's being deprecated now.
I believe that shaving a few additional kb from this already low number isn't worth the proposed complexity.
But this is a tradeoff that everybody has to decide for themselves.
If that's the argument, why not just let them take the (tiny) initial 8kb hit as an external css file, and make the rest of the experience even "zippier"?
One reason is Google will likely ding your Core Web Vitals score for it, which is becoming a ranking signal in May this year (https://developers.google.com/search/blog/2020/11/timing-for...) so you have less of a choice here for what you prioritise.
But haven't tested that yet.
The vast majority of users will visit one and only one page.
Have you considered inlining CSS in the head (as you've done), but then serving it again with a linked css file just before </body>.
Then, with subsequent page loads (of the current or other pages), you don't have to inline any CSS anymore.
Of course this requires that you serve two versions of all your pages, one for if that page is a first hit, and another to be served to users who already have your CSS cached.