That's pretty much the point: You don't have to remember what the function does, just read its name (and function signature, if your language is statically typed).
This is a "best practice" in imperative land as well.
In imperative land, I find that I have to write fewer functions, for better or for worse. As a result, this is not as much of an issue.
Not a problem in Haskell and Idris, at least. With the :search REPL command it will give you all the functions that match the signature you give it as an argument.
And really, you can have "longer functions" in any functional language, and they're used all the time: let..in...where, piping transforms (with the |> pipe operator or the . function composition), and of course, the do-notation which resembles imperative programming while being just syntax sugar for nested monadic operations.
Admittedly this is mostly a personal organization issue, but when writing object-oriented code, in my experience it encourages better encapsulation. That being said, I like functional more. :)