I usually try to stay away from these file vs search discussions; everyone has his/her view and usually are pretty set in their ways. While I generally side with the Calibre dev on this issue (his party), I don't actually use Calibre to maintain my ebook library because I do prefer my own folder structure. I generally import a couple of books I want to have on the reader into Calibre, convert them to LRF/EPUB, delete the source formats from Calibre, and load the converted books onto the reader. So you could say I am slowly building a LRF/EPUB library next to my original one. I don't have much of an issue with this; disk space is cheap and plentiful.
Since Kovid asked for a use case, here is mine. This is the top-level folder of my library:

(please forgive the rather basic presentation, mc rocks in everything but looks)
This top-level organisation has been fairly static; I've had it (with only minor changes) since the early 90s (talk about being set in my ways, huh?). Below it you will generally find a second level of folders (this is an example from the 'Computer' top-level folder), these are generally more fluid and prone to change over time:
Finding a book is a matter of using a file manager to go into the top-level category and then a second-level category, then executing a search on a certain keyword (name, author, whatever), searching though an already vastly restricted subset of books, speeding up my search by a significant amount. The need for this becomes especially apparent when you consider the size of the top-level folders (again, collecting electronic text since the early 90s):
Searching through the entire collection takes quite a while! I could replicate all this in Calibre, replacing folder names with keywords/tags but the task seems rather daunting. In addition I don't know if Calibre could handle this amount of files, I once loaded some
11000 project Gutenberg files (the LRF torrent of which is still going strong, by the way) and while still usable, the performance hit was very noticeable.