An extremely well-done module, worth both using and studying. For most practical purposes, you will not need to look further. Even if you do have special requirements, the best way to start is to study Steffen's work. Steffen is clearly a skilled practitioner from the old school of C programmers.
UPDATE: I've now put out a non-developer's version, so as of this version and date (0.206000 and 31 March 2008) Parse::Marpa is fair game for reviewers. I hope they'll bear in mind that current releases only claim to be alpha in quality.
I was surprised to see my first review, as I've put out nothing but development releases and, as the reviewer candidly, if ungently, notes, they are not at this point useful or documented.
Marpa is a combination of LR(0) and Earley's, while Parse::Earley is a straight implementation of Earley's algorithm. A glance at the code will show that Marpa is a much different algorithm, one that may be thought of as a combination of Parse::Yapp and Parse::Earley. Whether this combines the best of both, or my reviewer's unkind prejudgment turns out to be prophetic, I don't at this point claim to know.
I should add Parse::Marpa was undertaken with consultation with the appropriate CPAN folks, and after a go-ahead from Allison Randal of the Perl 6 team. It's our hope Parse::Marpa will be a useful part of that effort. None of which is to say the result won't be "garbage", but I think that most seasoned Perl hands will agree that it is prudent to delay that decision until a released version is available.