Production Server Nginx/Gunicorn Configuration Issue

My Gunicorn workers aren't handling requests efficiently under load -- I'm seeing ~30% CPU but high latency and a queue of pending connections in nginx upstream buffers. Current config: workers = 4, worker_class = sync (gunicorn), reverse proxy with keepalive_requests 100. Should I switch to uvloop/gevent workers, or tune the listen backlog on both sides?