Researchers comb search logs to learn where users get stuck

I find this very interesting. I think personally I tend to do two kind of searches - ones that I know in advance will only take one search, either because I am using the search as a lazy resolver (say because I type "BBC" knowing that google will return me a link to news.bbc.co.uk near the top, thus saving me a bunch of typing) or because the results will steer me to a definitive site (eg, if I type "tethering" I expect the Wikipedia page will crop up high, and I can figure out what it is [aside: why did we need a new term for modem?]).

The second kind of searches I seemed to do are the ones that really go quite deep, either because the information is hard to get, or because it is complex enough that I need to combine data from different sources, or because, as they say, I have phrased my query poorly and need to iterate to figure out what the magic words are.

I would be interested to know whether this is typical (ie the ditribution is bi-modal as I described) or just of descending frequency (ie most searches end in 1 query, the next most frequent searches end in 2, then 3 and so on)