I think that the compilation flags utilized are needed in order to properly justify the validity of the microbenchmark here.

At opt-level 2 and 3 rust completely optimizes away the interior loop, leaving just two instructions (plus a check for if n=0). given that your graph isnt showing O(1) time complexity shows you either

1) didn't compile your rust code in release mode for benchmarking, as the release profile uses opt-level=3 or 2) intentionally downgraded the opt-level of the release profile in order to run these tests

I'd be interested to see a more proper example, one where the task itself has some non-zero time complexity. Show a quicksort implementation perhaps. But as it stands this is just marketing smoke and mirrors.