WSL1 -> WSL2 wasn't only about filesystem performance. It looks like the Fuschia people are going to relearn those lessons the hard way. Namely, that it is much less work to make a real Linux kernel work well as a virtual guest than to reimplement compatible Linux kernel semantics in a host-OS-specific compatibility layer, and the former has better performance too. Also a lot of the work for the former is shared with other vendors whereas for the latter you're on your own.
What about Illumos's LX-branded zones? Bryan Cantrill has bragged about how complete that implementation is, about how it could even run something as exotic as the 100-language quine relay [1]. Sure, Joyent (who did most of the recent-ish work on this) ended up exiting the public cloud business, but surely that had more to do with the brutal winner-take-all nature of that space than any failure of the technology.