If someone was going to deploy a version of the Linux kernel with SoftBound (which is among other things what this paper suggests), they would have done so by now. It hasn't happened, because (a) contrary to what that paper suggests, performance really is a concern; (b) the entire ecosystem of C is what makes it so popular, not the language itself. Replacing the runtime is a huge amount of work.

C performance is a result of 40 years taking advantage of UB optimizations.

I think they were talking about SoftBound vs vanilla C performance, not Rust vs C.