Great questions! Anonymous Monk is correct in pasting my copyright notice, all the code for RPerl is copyright in my name and made available under the Artistic and GPL licenses, just like Perl itself. This means that a company can use the software without paying for it, but a company can NOT re-sell RPerl without the source code as they could under a BSD-type license. So wherever you use Perl you can use RPerl.
You are correct about being able to combine low-magic RPerl with high-magic Perl. The 10% of your code could be written using RPerl and give you a speed boost.
The current state of RPerl development is available for anyone to see on Github. I've created typemaps for the basic scalar, array, and hash data types, and I've hand-compiled a sorting algorithm into both RPerl's Perl-data mode and RPerl's C-data mode, which is what gives us our legitimate performance benchmarks. I'm working now on the automated compiling of said sorting algorithm, which is part of RPerl v1.0.
Please do not hesitate to ask me more questions, I am very interested in communicating with potential RPerl users such as yourself. Thanks!