To give context, the implementations for the same problem are completely different, e.g. javascript versions uses worker thrreads on 4 cores for most of the examples. To compare actual performance first similar approaches should be used.
Spectral norm is an interesting example because on single thread and without using SIMD, C, C++, Java, Javascript, Swift and Dart shows very similar performance characteristics (5-6 seconds). The loops are tight, there is not much to do for a decent compiler. it almost pure simple 64 bit floating point arithmetic.
Then you add multi-threading, and see a decent improvement, e.g. from 5 seconds to 1.5 seconds on 4 cores (almost all implementations show a similar gain)
And some languages with support for Vector operations further improves this to 0.7s
So this benchmark seems to be not very useful for comparing performance of any decent language. It could however be used for comparing improvements of using different techniques to solve the problem, or to showcase what new features brings to the table.
