That's a fair point.

I don't quite follow why speed needs to be sacrificed to get predictable performance though.

And I don't know about writing a (whole new?!) Rust compiler (!) - if the current Rust compiler wouldn't be a good fit for this purpose, well, I definitely don't have the resources to anything about that, considering the unbelievable amount of attention to detail and wrangling that's gone into getting rustc to where it is today.