If they increased the heap until the JVM was exactly as fast as their compiled Rust code on the same/equivalent hardware, you could get those numbers as some ki

If they increased the heap until the JVM was exactly as fast as their compiled Rust code on the same/equivalent hardware, you could get those numbers as some kinds of GC significantly decrease overhead with increased heap size. With that said, that's an almost entirely useless benchmark, except as a testament to the amazing JIT compilation the JVM does. That the JVM with GC is capable of matching a compiled language specifically made to have efficient memory management at ANY heap size is positively astonishing!