This isn't clear from the blog post, but we only use autocommit for read-only queries. Using autocommit is only really a major boost if you have a very high skew of reads to writes like we do (50:1).
Thank you for clarifying. Do you segregate autocommit based on whole chunks of application code (for example, front-end web servers are totally read-only at the connection level and therefore autocommit is fine) or is it more ad-hoc?
