Set mv_substring_match to yes (su=yes in one-clicks).
This most commonly happens when searching for non-ISO-8859 (Cyrillic, or characters like umlaut and eacute) characters in word-match mode. The problem is, that unless your locale is set up properly that Perl doesn't think a non-ISO-8859 and a space character is a boundary.
Also, if you are searching for non-alpha characters then they will also not be interpreted as word characters and the boundary problems will still exist.