In a multithreaded context, shared data was shared via atomic reference counting. The atomic operation acted as a single bottleneck for every operation. The sol

In a multithreaded context, shared data was shared via atomic reference counting. The atomic operation acted as a single bottleneck for every operation. The solution was to derive deep cloning for the shared data (a 1 line fix) to reduce atomic contention.