I guess part Hyrum’s law in that big projects depend on implementation details to a degree, related but many C extensions (though Graal can run these as well, but perhaps not all of them), and last I read about it, the JIT compiler is not yet that good at optimizing very big applications — it takes a really long time for some functions to become hot enough.

I think it's just that the very largest codebases are the most likely to end up depending on obscure bugs in Ruby or the surrounding ecosystem itself, or maybe features that basically nothing ever uses except that one file buried deep in that one gem that three real programs use. Ruby apparently has a lot of stuff like that.