If you've ever looked at raw search logs and tried to tally them you know how noisy that data can be. You have to filter out all the bogus robot hits, normalize per source & partner (think a static href link into a search, like piggly wiggly appearing on a high traffic page. So do you de-dup by referrer as well?) You start throwing out the edgy stuff and playing with stripping out quotes and coalescing typos and misspellings and before you know it you can have whatever list you want when you're done.
At Netscape I found that the code used to tally the reports actually discarded the search that more people did than any other search. It was top by a huge margin too. No not sex, yahoo, porn, britney or anything like that.
It was the null search. People just clicking on the search box without having typed anything. Of course the empty string was stripped out during the log tally run...but inadvertently discarding a key piece of user behavior data.
We had an error page which came up when that happened...we quickly replaced it with a friendly page of instructions. Doesn't seem like Yahoo and Google do anything at all with that search now, MSN does a page turn onto a null result page.