Comment by lfborjas

Check out rubinius, the bytecode seems to perform considerably faster than interpreted source: http://rubini.us/