Hmmm.
The FusionAuth basic tier is not intended for any type of load testing. So that probably explains poor results.
The reason for this is that the service (with this tier) is running on a single AWS EC2 instance that is running FusionAuth, Elasticsearch and a PostgreSQL database. This means the node is very resource constrained, and attempting to run load tests on this type of system is not recommended. The numbers you get from this type of test will not be valuable to you in context of planning your production deployment.
If you do want to perform load testing, I would recommend you run FusionAuth on-premise using a more realistic production configuration, or upgrade to the High Availability hosting tier. You can spin up an HA instance, perform your load testing, and then tear it down. You'll only be charged for the time the HA instance is up and running.
If you cannot achieve your target request per second with a standard HA setup, you can upgrade from Medium, to Large, or even to X-Large if you need to obtain additional requests per second.
You also may be interested in this load testing guidance: https://fusionauth.io/docs/v1/tech/installation-guide/monitor/#load-testing
And this (dated) forum post: https://fusionauth.io/community/forum/topic/8/what-level-of-performance-can-we-expect-with-using-fusionauth-as-an-idp