Say you memorize a random 16 character password. With 16 characters from a set of 95, there are a total of 9516 possible combinations. Using two layers of encryption each with an 8 character password, each of which has 958 combinations, the maximum number of guesses an attacker would need is not the expected 958 × 958 (9516) but 958 + 958 (959), which is much, much lower. Because of this, one long password would give you a security level of 9516, whereas two half-sized passwords would only give you a security level of 959. The attacker would only need to crack one container at a time.
The only time that using multiple layers would be useful is if you want to use different passwords in order to prevent compromise of the disk encryption password from implying compromise of the container. Because computers naturally need to keep the encryption key in memory, a cold boot attack or other physical attacks may be able to retrieve the key if the computer is powered on. By using a separate container that you close whenever you are not using it, you minimize the amount of time the container's password is present in memory without having to sacrifice full disk encryption.
So, what should you do? Use a single password created with the diceware technique. That involves using physical dice to pick out words from a 7776 (65, the number of possible combinations of five 6-sided dice throws) word dictionary and using the space-delimited words as your password. The amount of entropy present in a diceware password is log2(7776n), where n is the number of words. To have more than 100 bits of password entropy, which is often considered the minimum to be secure for the foreseeable future, you will need to have at least 8 random diceware words. Any less than 6 words and you start getting into dangerously weak territory where cracking becomes possible.