The performances of KNet exceed the one of Confluent.Kafka. Is it possible?

Side-by-side the development of KNet a benchmark test program was developed too.
Several tests were made and a first report version is available on performance page.
From our tests (see the report for the approach used, configurations and so on) emerge that KNet works better than Confluent.Kafka in many conditions.
This is a great news for development team, anyway maybe there is something made wrong in our use of Confluent.Kafka: I mean API misuse, wrong configurations or other.
If someone get a look to the benchmark test program, available under tests folder of its GitHub repository, and found something wrong it can be useful.
For newbees of KNet just a simple description of it: KNet reproduces in .NET the same API surface of Java version of Apache Kakfa; to make its jobs it interfaces with an in-process JVM so, more or less, it is like a Java client usable from .NET.
If the performance are confirmed, KNet can be a good replacement of Confluent.Kafka: from results it is performant, it is designed to implicitly evolves each time the Apache Kafka Java version evolves, simple to use because shares the same API surface of the official Apache Kafka Java version.