Github has got a bit of a read mostly situation. If read requests where funneled so that there was only one per URL per edge node at any given point in time (which edge side includes effectively gives you), it would make for pretty easy to manage load for Github.