Padraic Brady has posted about an issue he noticed when working with regular expressions and the "word" character type to find something that's alpha-numeric (including an underscore):

You can find the "word" generic character type used in a lot of PHP code including the Zend Framework. The problem is that the assumption above is incorrect. Now, most of the time these act identically because PHP is compiled using its own packaged PCRE library. However, I've seen more than once systems where this is not the case. Usually in some non-English capacity where additional locale support was considered necessary or standard practice.

The problem comes when PHP is compiled against a custom PCRE library, making it more locale-aware. He gives instructions on how to get this to a testable state on your environment (using an updated PREC library) and get it working for characters in French, like the accented "a" or "e".