Prompted by this post I tried to run my latest Rails 4.1 app with Rubinius. It's been a long time since I wished to give it a try. I had to comment out some gems which don't work with Rubinius, all stuff I use only in development (byebug, binding_of_caller, did_you_mean). There is a Rubinius debugger but I didn't try it http://rubini.us/doc/en/tools/debugger/ I had to remove the space after the arrow in lamba definit…
On the other hand Rubinius implements native threads, which means for parallelizable workloads you get a potentially equivalent performance gain to the regression you are talking about. Currently the only way you can get this is JRuby (which of course comes with own its own JVM pros/cons). I'm not using rubinius yet either, but I think there is already a compelling case to take a serious look at it.
Re: Rubinius 3.0 – Part 1: The Rubinius Team
#11"native thread" is just buzz word if the implementation cannot boot rails. On the other hand, if you don't feel like to use real multithread now, it's unlikely you need it in the future.