main.dat includes global/*.dat, and then by-date/*.dat.
by-date contains 2012/ and 2012.dat (and another such pair for each year, when I've finished entering older years), which includes by-date/2012/*.dat (which are monthly).
accounts.dat is all of the accounts I deal with, and defaults.dat is for stuff like setting the default currency.
That puts a nice limit on the size of each file, and I find it superior to dividing things up based on accounts because that can be icky when entries involve multiple accounts. In order to classify entries, I just use tags. It also makes importing from a monthly CSV even easier, since it can just go in its own file.