This is due to the use of isalpha() internally, which doesn't play well with multibyte encodings like UTF-8, regardless of the locale setting.

Fundamentally, this is the same issue as bug #27668 — I'm not sure there's a lot we can do about this in PHP 5.x, but it's worth noting if and when we revisit Unicode string handling internally.

Thanks for the reply. Given your comments about the problems, would it be helpful 
for me to also file a feature request for newer versions of php to have a 
mb_str_word_count function which could properly handle this case? I haven't dug 
into the C code enough to understand why isalpha() fails on multibyte, but I'd 
have to imagine there is an alternative available that will handle multi-byte 
characters properly. I could potentially even create a patch if pointed in the 
right direction.

Yeah, a feature request for mb_str_word_count() might be a good idea.

The isalpha() issue isn't really PHP specific: the underlying C function simply takes a single byte as its input, so it can't ascertain whether a multibyte character is actually alphanumeric or not (since it only ever gets the first byte of the sequence). There's an iswalpha() function that would do the right thing, but PHP was written before it was widely available, and using it in str_word_count() alone would be inconsistent with the rest of the language: it's something we'd need to think about as part of making the whole language more multibyte-aware.

Ok feature request filed here: https://bugs.php.net/bug.php?id=63671
First time doing that, so hopefully it's correctly filed.

> […] it's something we'd need to think about as part of making
> the whole language more multibyte-aware.

That's unlikely going to happen, so I'm changing to doc bug.