I'm not someone on the rust bandwagon, but I doubt it's a matter of performance as much as safety.

Remember Cloudbleed? That was caused by a memory safety issue in a parser written in C++.

re: jvm -- you may be able to get that sort of throughput, but I also have to wonder what the latency distribution looks like, and how much control you wind up having over it. Yielding control to a GC in a latency sensitive application isn't an easy give.