I saw that the performance tests were made with Amazon Corretto JDK. Their JDK differs from the offical OpenJDK and can include performance optimizations.
Thats why it would be interesting to run the benchmark with OpenJDK and to compare the results with Amazon Corretto.