By definition, urlencode() should leave alphanumeric characters unencoded. However, (uin my case) it doesn't respect locale settings, and encodes non-ASCII (but alphanumeric as per locale definition) characters when it shouldn't. Locale is correctly set up, and Apache does have correct LANG and LC_ALL environment variables in its runtime environment (checked by phpinfo()). I even tried executing explicit setlocale() call within a script but this doesn't change anything (which is probably correct as locale is already set systemwide)

I don't know if this problem is peculiar to FreeBSD installations, perhaps someone should check this out - might be possible, since judging by source code, system isalphanum() is used to determine whether a character should be encoded. However, FreeBSD in general handles locale very responsibly and this wouldn't be possible if fundamental checks like isalpha() were broken.