Yeah, this error is made consistently throughout the article and the author should fix it. It serves to inflate the project's performance by as much as two-thirds, and some people will see this as intentionally misleading and write your project off because of it. ultrajson looks to be way faster; don't alienate people by fudging the numbers.