Your analysis is correct in this: randomness is everything. The attacker will try "potential passwords" and will be defeated only by choosing your password in a large enough set, so that the number of "potential passwords" far exceeds what the attacker may practically try with the computing power and free time available to him.
In practice, repeating the same word several times is a classic trick, which many people use because of the widespread myth that longer passwords are stronger: this is wrong, but many people nonetheless believe it. Since many people follow the repeated-word strategy, password cracking tools follow it, too.
Therefore, the best you could hope for the entropy of a repeated-word password is the sum of the entropy of the base word, and the entropy of the repeat count (or total password length)(this is a sum because I am expressing the entropy in bits, which is a logarithmic scale). For instance, if you choose the base word in a list of a bit more than 30000 possible words (which is an already quite large vocabulary, by today's standards), then that's 15 bits for the base word. If you then repeat the word to achieve any length between 12 and 27 characters, with the length being chosen randomly, then that's 4 extra bits (16 choices for the length), for a grand total of 19 bits of entropy -- i.e. not a lot of entropy after all.
An extra point to consider is that password length may leak. Not in the hash, but in other contexts:
"Shoulder surfers" may have a peek at your screen when you type the password. They will only see a collection of bullets (password entry fields are hidden), but they could see how many such bullets appear, yielding the password length.
In a similar context, people within earshot may try to work out the number of keystrokes, there again yielding the number of passwords.
In an HTTPS context (i.e. a login form of the Web), the password you type will be sent into the SSL tunnel, encapsulated in a Web form. Eavesdropper on the line will not be able to see the password (SSL does encryption) but they will be able to see the length of the HTTP request, because data length leaks from SSL (with single byte accuracy when an RC4-based cipher suite is used, as Web servers are prone to enforce in a slightly misguiding attempt at defeating the BEAST attack -- which does not work anymore, by the way).
Summary: the repeated-word strategy does not yield a lot of extra security, possibly none at all if the password length has leaked, which happens in several password usage contexts. As such, this strategy is not a good usage of the user's willingness to type a complex or long password (and that is a scarce resource).