The MediaWiki search defaults to excluding 547 words as stopwords. But they're perfectly good words (you can see them on the searchtools site). It's a MySQL full-text search default, and the MediaWiki people have never changed it. Exactly like the short words in the previous rant, these words are not indexed at all, so can never be retrieved by the search engine. Stop words include: able, about, above, according, across, actually, after... So a site search containing only one or more of those words has "No page text matches", even when there are pages with those words.
Example at knoppix.net, tried the seven stopwords above, not one match
This message is not just unhelpful, it's misleading. It doesn't even say which of the search terms are stop words, so there's no way to tell except trial and error (or looking at the list). But, contrary to the message, specifying a search with an allowed word and a stopword or two, such as surprise from behind will match all articles containing the word surprise, without checking that the article also includes from and behind. Whoops.
There's a wikimedia meta help page with the awkward title of, Common words, searching for which is not possible. I find this all pretty user-hostile, and I think it stinks.
The main Wikipedia removed stopwords from search in February 2006. They don't say exactly why, though I find it blindingly obvious. But the MediaWiki installation still uses the giant stopword list. To fix it, reconfigure MySQL, or try the procedures some nice user has posted. Reduce the stopwords list to reasonable minimum (the, a, an, and, or, not), or leave it out altogether. Or switch to Sphinx or MWSearch (Lucene) which have fewer stopwords by default and can be set exclude none at all.