Comment 5 for bug 1031954

On my (new) Macbook Pro (specs in that gist), a Python 2.7-pickled part-power=18 ring took ~3271ms, while the same ring pickled by Python 2.6 deserializes in ~6ms. That's between 2 and 3 orders of magnitude (545x).

You should be able to run the same benchmark script (dependencies include having a ring in the current directory named object.ring.gz and the benchmark module installed) with a RAX-production-cluster-size ring and see what that looks like.