So I'm a bit confused, the paper says "'Randen' ... outperforming ... PCG" but now you wrote "though slower than pcg64"? In the paper itself, PCG is faster but "these microbenchmark results are quite irrelevant in practice — which actual application repeatedly calls a random generator and ignores the results?" I wonder, was the paper peer-reviewed? (I'm not saying it should have been..., just wondering.)
My guess is that yes, Randen is fast, but it is not quite as fast as PCG or other pseudo-random number generators.
Whether or not it is relevant, that's another story. I can not agree to your claim "there are vanishingly few use cases" for insecure random number generators.
If you look inside Abseil it seems that the authors generally agree on performance. The default random in Abseil is Randen, but the faster variant the use of which is discouraged is PCG.
The paper doesn't seem peer reviewed or published. It doesn't show up as published anywhere on Google scholar, and papers that are published years later still reference the arxiv link, not a publication link. This is pretty strong evidence of not being reviewed or published.
You can check via: google scholar, search randen random number, note there is no version with a journal, click citations and look through the few articles mentioning it, you can see which of those are published, can see a few, not one has a journal ref to it.
I was citing the micro benchmarks because they provide a pessimistic comparison in favor of other RNGs. My recollection is that it appeared as a conference paper, but I'd need to ask the author.
> the micro benchmarks because they provide a pessimistic comparison in favor of other RNGs
Why would one claim they're fast, then simply fiddle with benchmarks so they find a place where they look fast?
When someone wants a fast RNG, they're not calling it 10 times per sec. They're hammering it for actual needs.