Naive micro-benchmark of reduce/sum I made to get the idea of raw performance range of different dynamic languages / different runtimes.
For each of these I repeated the execution until the timing stopped changing in major way (JIT etc), and saved only the last run.
Don't take it too seriously ;)