In general, attackers are adaptive. They know what people think. If users tend to begin their passwords with 'z' then attackers will start their brute force with that letter.
Any specific strategy, such as choosing a 'z' as first letter, may give you an edge over the attacker only as long as the attacker does not know it; so talking about it on a public Q&A site is not the smartest move ever. Moreover, attackers always have the option to try passwords in a random order, which ensures a non-worst case for the attacker. When there are N equiprobable passwords, the average cost for the attacker is N/2 tries, and by trying the passwords in a random order, the attacker achieve that cost regardless of your choosing strategy.
The important point is that password protection is stochastic. For any specific password instance, an attacker can get lucky; he can also be especially unlucky. We thus analyse password resistance based on the password generation process, not based on a specific password instance: the generation process tells us what will be the average success rate of the attacker. "Average" is the crucial word.
Take note, too, that probabilities rule your life: whenever you get out of your home, you are taking the probabilistic bet that you will not be struck by lightning or bitten by an rampaging Rottweiler. You accept this because you instinctively perform a cost/benefit analysis and consider that such events are sufficiently improbable that they don't outweigh the benefits of being able to walk outside.
That concept just applies to passwords as well: you will rely on your password generation process having sufficiently high entropy, i.e. making attacker's success rate sufficiently low, that attackers will not even try (or won't succeed with high enough probability). In that sense, "aaaaaaaa" is not stronger or weaker than "zzzzzzzz", as long as the method you use to generate your password can produce both with equal probabilities. The strength of a password is not in what it is, but in what it could have been.