It's for Lua, but there isn't anything in the VM design that requires Lua per se - porting it to another language's C API wouldn't be difficult. In my experience, LPEG has been quite easy to use for the sort of tasks one would normally apply regular expressions to (and then some, as it handles recursive structures far better), it's quite fast, and it's easy to tune incrementally. In my (not yet ready for release) Lua/LPEG webserver, I'm handling 1000+ requests per second in about 2MB ram total* , so LPEG's design has tamed the usual space issues caused by PEG memoization.
* Only a tiny amount of that time is due to request parsing, it's mostly because I'm still working the kinks out of the event loop.
One strike against it is that there isn't a whole lot of casual documentation, though the documentation that does exist (the research paper and a library reference) is quite thorough. Also, it takes advantage of Lua's operator overloading to make the grammars concise, which takes getting used to. Still, having it so thoroughly integrated into the language is extremely convenient, and it's quite a bit more expressive than regular expressions.
Also, here's a link to the PEG paper on the author's site, rather than behind the ACM's paywall: http://www.bford.info/pub/lang/peg.pdf . Gotta love the ACM.
Grammar ambiguities are resolved by ordered choice. If there are two ways to parse something, the first way will be tried first. If it succeeds, the ordered choice short-circuits and the second way is ignored. So you can resolve, say, the classic dangling else problem by ordering your choices correctly.
http://compilers.iecc.com/comparch/article/05-09-114
In particular:
"Yes, this is "the" problem with PEG's. / implies ordering of the search (parsing) space. You need to order your / operators so that special cases (e.g. longer matches), so that they appear first. Unfortunately, if you don't do this, nothing will tell you you have a problem with your grammar, it will simply not parse some inputs. To me this implies that if one wants to use a PEG to parse some input, then one must exhaustively test the parser."
Hopefully the thread therein will describe my issue better than I have thus far.
While the backtracking behavior is different, choice order affects PEG rules similarly to Prolog.