Array traversal magnitudes faster after removing the first element

Question

After reading the recent smashing magazine article on optimisation, I ran some tests to see what would be the most effective way of "deleting" an element from the middle of one of my arrays.

After running my own tests regarding splicing a value out from the middle of the array vs deleting it/setting it to null, I came across the rather unexpected result that splicing the value out from the array made the array an order of magnitude faster to traverse.

More investigation led me to this.

For some reason, .shift()'ing the first record from the array made it 300 times faster to traverse (the biggest performance being seen in v8, but it seems to be valid for all the browsers I've tried it in).

I doubt I'll abuse this as I don't think the actual traversal is a bottleneck, but does anyone know why this behaviour occurs?

Edit: Incorrect use of jsPerf was the underlying issue here, see my answer below.

Answer 1

The performance test is filled with countless errors.
Most important one is to take a sample length of 1000. With today's processor, traversing a 1000 items array is done instantly, and you measure more the time taken by the first array method you call (shift, slice, ...) than the actual array traversal.
So you have to
1) use a much longer array,
2) do the slice/shift/... BEFORE the loops since that's not what you want to measure.

You will then see that there's no magic, and that the array traversal takes the same time for all arrays.

i started in http://jsperf.com/spliced-vs-non-spliced/4, for normal/sliced(0,0) and shift() cases. differences in performance are inferior to measure error.