I can't remember if it was /The Economist/ or BBC, but I remember
reading somewhere a statistic that showed that most people relied on a
external search engine to find information instead of starting from the
head of a site and working inwards. Sometimes the latter saves several
steps, though...
The reason is Google sorts hits by how many links have been created to
an entry. Usually what you want is near the top. The onsite searches
almost always give you a useless hodgepodge of irrelevant trivia
sorted in no particular order.
The problem with tree structured menus is authors like to arrange only
ONE way to get to something. My thinking is you should try to cover
pretty well any plausible route, with lots of cross linking at the
lowest level to other similar things.
I originally wrote the Java glossary for my own use. Its various
searching mechanisms I have developed over time to make it easy for ME
to find stuff. If every site's authors used their own indexing, I
think the quality would improve rapidly.
It takes work, but creating manual index to the critical information
is always going to beat a mechanically generated one that just finds a
word used, no matter what the context.