I pretty don\'t understand why this test :
http://jsperf.com/push-method-vs-setting-via-key
Shows that
a.push(Math.random());
Could you explain why this is the case?
Because your test is flawed. The push does always append to the existing a array making it much larger, while the second test does only use the first 1000 indices.
Using the setup is not enough here, you would have to reset the a array before every for-loop: http://jsperf.com/push-method-vs-setting-via-key/3.
Apart from that, the method call to push might have a little overhead, and determining the current array length might need additional time compared to using the index of the for-loop.
Usually there is no reason not to use push - the method is there for exactly that operation and makes some code easier to read. While a few people think one version is faster than the other, both are equally optimized in browsers. See Why is array.push sometimes faster than array[n] = value? and Using the push method or .length when adding to array? - results vary so wide that it's actually irrelevant. Use what is better to understand.