Would it make sense to submit a rails-optimized pull request for this benchmark that replaces some key performance bottlenecks with appropriate C gems? I'd be curious to see how fast rails can go out of the box without doing your own hand optimization.