Wow, direct assignment is an order of magnitude faster than push (in my test, chrome 27 on windows 7). Can anyone explain why?
Direct assignment is the only test in which the length of array doesn't change.
this smells like a js engine optimization. you wouldn't expect substantially different machine code to run for "Direct assignment" and "other". i wouldn't be surprised if the loop is being unrolled or something.
They could all be reasonably unrolled, so it's probably something else.
I expect the engine can more easily calculate the final required Array size and preallocate it upfront avoiding resize operation on each loop iteration. (which is anyway JS best practice when possible)