Most of the problem lies in Database. Rails may not be best architecture for scale. But I doubt you could even get 100x difference if the bottleneck is in Database.
I don't know any large, JVM based WebSite in large scale on top of my head, but I consider Stackoverflow, written in ASP.NET to be one of the best and most optimised site. Near 700M Pageview per month, with 10 Front End Servers. At peak it does close to 5000 RPS, Cookpad does 15,000 RPS with 300 Rails Server. But the SO servers are at least twice as powerful, so that scale is like 500 RPS / Server to 100RPS / Server. 5x Difference.
Does SO probably has a lot more cacheable content and their core content size is small < 2TB full DB size with DB memory hitting 768GB or so. Not that it doesn't require good engineering. I do love their stats.
You're making the assumption that there is a database in the mix for the 100X case. There wasn't, except in-memory. It wasn't a 100X improvement across the board, it was 10X to 100X.
Did Twitter ever upgrade past Ruby 1.8? If not these numbers won't be particularly relevant to modern Ruby. (for anyone else reading 1.8 was a simple interpreter and 1.9+ a modern high-performance bytecode VM)
Most of the problem lies in Database. Rails may not be best architecture for scale. But I doubt you could even get 100x difference if the bottleneck is in Database.
I don't know any large, JVM based WebSite in large scale on top of my head, but I consider Stackoverflow, written in ASP.NET to be one of the best and most optimised site. Near 700M Pageview per month, with 10 Front End Servers. At peak it does close to 5000 RPS, Cookpad does 15,000 RPS with 300 Rails Server. But the SO servers are at least twice as powerful, so that scale is like 500 RPS / Server to 100RPS / Server. 5x Difference.