I've finally looked up
Brain Wallet, and your scheme is essentially the same: the passphrase is used to generate or recover the private address through some complex mathematical functions. The "brain wallet" described in the link simply uses the SHA256 hash function to generate the private key based on a passphrase. Your "encrypted private key" uses a passphrase to scramble the private key, but you also leave a clue to help you remember the passphrase.
Essentially, in both schemes, the attacker wants to generate your private key, and to do that, he will need to guess the passphrase and perform the proper mathematical operations that would generate the private key. In the "brain wallet" scheme, the mathematical operations are well know; that makes it easier for multiple pieces of software to be written that can generate the private key with the owner not being required to know the details of the math functions. You are using a non-standard order of mathematical functions to generate your private key, but you are also giving the attacker information that will help him guess your passphrase, and if you do end up using the encryption procedure you've described, he also knows the math functions, so you're scheme is no more secure than the "brain wallet" plus you've given clues to your passphrase.
This thread is really a discussion about "how do I come up with a good passphrase with lots of entropy."
One alternative to coming up with a good passphrase might be to use a bad passphrase, such as "Mark Twain," but run that through the MD5 hashing function then use that output as your "brain wallet" passphrase. An attacker assuming that your "brain wallet" was generated with a simple or complex passphrase would put things that look nothing like an MD5 result into the SHA256 function, and you would have an effectively good passphrase, but if the attacker knew your procedure (or used more sophisticated methods than brute-force), he would probably get your private key in short order.