The problem with many of those articles is that it's usually not "idiomatic implementation in X" vs "idiomatic implementation in Y", but more like "yeah, X was better, so spent hours optimizing and profiling the hell out of Y, it's barely readable but it's 30% faster!"
It is a benchmark of seconds and the startup of the entire JVM is included in the benchmark? And the Java implementation is still faster? Yeah something is definitely wrong with your implementation then.
Since it's a benchmark about an algorithm why is the JVM startup time even included in the first place?