I'm not sure how relevant this is, although heaven knows getting decent performance out of EBS is a worthy goal.

The trouble is that the numbers he's seeing at the high end have to be coming out of ram cache. This effect of "sometimes the my benchmark was served out of cache more than others" completely dominates the results and without WAY more testing to average things out the numbers are worthless.

Same post, different URL, but with a lot more comments & discussion: http://orion.heroku.com/past/2009/7/29/io_performance_on_ebs...

(bah!)