I'm not sure where you get that from? The benchmark code clones one read handle for each read thread at the beginning, which is then used for the entire benchmark: https://github.com/jonhoo/rust-evmap/blob/d307999c1ad78d10ec...