A comparison to perl 5 + XS shows a factor of 130, which is *much* closer that the 350 from August, and that is speed improvement in the language itself.
In theory, Rakudo is a language with a better type system than C. In theory, Rakudo has a JIT and shouldn't require you to write XS or C or internals code to optimize things. In theory, a better optimized system wouldn't have to cross the language/ABI/calling conventions boundary of Perl/XS.
In theory, Rakudo was designed to be easier to optimize (and to require fewer pessimizations) than Perl.
Now I know as well as anyone that optimizations are difficult and time consuming and need a lot of careful thought and tuning, but parsing and file I/O aren't exotic operations that tax your ability to think of complicated optimization strategies. Rakudo/Moar still being over two orders of magnitude slower than optimized Perl for CSV parsing is not, to me, a ringing endorsement.