Benchmarks aren't, and never will be, the reason for using Ruby on Rails. Quite a bit would have to change in Ruby the language in order for the detractors of Rails to change their minds, and lets be honest, Rails wasn't built for those type of people anyway.
I agree. Ruby is slow for certain types of workloads, but for many kinds of workloads it simply doesn't matter, and Ruby is fast enough.
For workloads that require super high performance, one can simply use something else... Java, Node.js or whatever. Now that Phusion Passenger supports Node.js as well, it's very easy to write some parts of an app in Rails, and write other (performance-critical parts) of an app on Node.js, and deploy them together on a single server with ease.
A C extension in Ruby can be nothing more than an algorithm implementation accepting strings or numbers and returning such. Where you need proficiency in C is when you're building something substantial in it.
The problem with Node is, you have to load up a runtime. That can not only kill your performance gains, but actually make things slower.
I'd say standardize on a deployment model that leverages lots of nuts and bolts ruby doesn't care about.
Specifically, try using the apr (apache portable runtime), or leverage tomcat (which can use apr) + jruby + jndi jdbc connection pools.
Let rails be rails, your best bet for performance is either making the underlying ruby runtime better or leveraging things that are designed to be high performance.
http://www.techempower.com/benchmarks/#section=data-r7&hw=i7...