That's really cool, excellent work. I'm almost surprised it became that much faster even in unoptimized form.
For optimum performance, standing on the shoulders of giants and creating an XS wrapper around lifelib would obviously be best, this would have the added advantage of supporting all (classes of) CAs that lifelib supports. That said, having a pure-Perl implementation that will run anywhere that Perl will is obviously valuable as well: lifelib is tied to a specific architecture and a handful of instruction sets. And to say you're beating the existing CPAN solutions would be an understatement.
(Speaking of CPAN: you do intend to eventually bundle this up and release it as a module, right?)