[pjmlp]

I would only accept such PR with profile data and benchmarks proving the point.

[ninkendo]

The dropbox article others and myself have linked provided such benchmarks, for an 11% speedup. Seem plausible?

[pjmlp]

As everything in IT, it depends.

That was for Dropbox specific use case, and 11% seems too much in any case.

For example, what improvements has had rustc in the meantime to better optimize bounds checking, it is still the same, it is less, does it actually matter on the acceptance criteria regarding the definition of done on the story.