"Array#each is rewritten in Ruby for better performance"[1] was a surprise for me, with a Ruby implementation being 5.3x faster than the C implementation[2]. YJIT seems magical.
Pretty much the story of .NETs CoreLib, including even the most performance sensitive parts like memcpy/memmove. There’s probably no C++ code left in anything that isn’t JIT, GC and lower level parts of VM.
Given how close modern .NET and C# have come to systems languages like D and Modula-3, among others, it would be interesting as language nerd, to eventually see it fully bootstrapped.
Probably won't happen though, for economical reasons, where to spend development budget.
[1] https://bugs.ruby-lang.org/issues/20182
[2] https://github.com/ruby/ruby/pull/6687