th_yc

Yes, since PyPy (and Rubinius) have JITs (like V8), that is the fairer comparison. Nevertheless this highlights just how helpful JITs can be for cpu-bound code -- makes me wish PyPy was more ubiquitous (and gets py3k support sooner rather than later).

Exactly -- for a fair comparision, V8 (which has a JIT) should be compared against Rubinius (for Ruby) and PyPy (for Python), which will both be nearly as fast as V8.

Of course, it is true that V8 is ubiquitous, whereas Rubinius and PyPy are not -- that is the one majore advantage of javascript.

Technically the above is a dynamic programming algorithm, in that you are avoiding the recomputation of fib(1..n-1) in each step. That is equivalent to recursion+memoization, which would perform roughly similar to the above. So it is not that "recursion" is so dramatically slower than "iteration" in general, it's just that for these kinds of computations, recursion should be memoized.