Link Details

Intriguing -- please see the ws3-1-Hundt.pdf document for all the interesting details.
There is more than the run-time measurements: the LOC figures are also important.
The java timings are not acceptable. So indeed Java is slow.
Also disappointing is GO: it should be faster, and shorter.

The java implementation use Lists instead of arrays, I think this is the main cause of the difference in performance and if this is the case the benchmark is unfari.
The document says that they use Lists because java doesn't support generics arrays but I think this choice is questionable.
The usage of non-generics array is possible with the addition of a few explicit casts.

Arguing that Java IS slow with just this benchmark seem wrong to me.

So what about the new parallel api in Scala ?

And i hope they compare Erlang, Scala and C++ OpenCL in concurrency performance

Good point, and I'd add that it's also questionable that they'd spend time "extensively tuning" the C++ version, while not even bothering to do a few casts w/arrays in Java. Detecting a hint of possible bias here...