Sasa Juric, and the folks at pusher wrote a couple of blog posts on this: http://theerlangelist.com/article/reducing_maximum_latency