Objects implemented with arrayrefs rather than hashrefs are often
faster than those implemented with hashrefs. Moose's default object
implementation is hashref based. Can we go faster?

Simply use MooseX::ArrayRef instead of use Moose, but note the
limitations in the section below.

The current implementation is mostly a proof of concept, but it does
mostly seem to work.