Job Stuck in Queue Eventhough worker is ready

I am using serverless endpoint with H100 but I am experiencing high queue time .If you send a single request to runpod enpoint you may get 2 seconds delay time and on same 2nd request you will get queue time of 7 seconds which should not happend.I think they should optimize their queue and worker communication codes
ist run: 3 seconds
2nd run: 15.84 seconds