Horrible scalability of my experiment

[Jiboxiake]

I have an experiment testing scalability of hash maps under a certain workload.
https://github.com/Jiboxiake/Concurrent-Hashmap-Experiments
In the Global.hpp I control how many worker threads (readers and writers and hashmaps) I have. Libcuckoo's hash map performs OK with fewer than 16 threads but works extremely bad with more than that. I'm wondering if this is a bug or I did something wrong. You can find my experiment in Libcuckoo_Benchmark.hpp.

Thanks.

[Jiboxiake]

The experiment result is here:
Libcuckoo Value.xlsx