If the attacker has knowledge of the associated bitcoin address, the problem is much like cracking a hashed password. Each test will require more compute power than a single round of a common hash function, but a lot less than 1024 rounds of MD5.
In the first one, your removed code letters specify the lower 32 bits of the encoded key, which are the checksum. To even create a private key like this, you would need to search for upper code letters that result in your desired checksum. The code letters in excess of those making up the checksum are the only part an attacker would need to guess.
The second requires the attacker to guess the code letters, as well as where they are located in the key.
The third requires the attacker to guess two groups of code letters and all permutations of where they might be located, which is a few more orders of magnitude harder.
This is basically correct apart from one thing: if you leave the checksum digits the hacker can reject the majority of their hacking attempts after just doing a single round of SHA256, which is much cheaper than having to compute the address or public key their guess maps to. Ideally you should remove the checksum totally and insert the vanity section somewhere in the rest of the private key, but that's kind of tricky to reverse. In the first one, your removed code letters specify the lower 32 bits of the encoded key, which are the checksum. To even create a private key like this, you would need to search for upper code letters that result in your desired checksum. The code letters in excess of those making up the checksum are the only part an attacker would need to guess.
The second requires the attacker to guess the code letters, as well as where they are located in the key.
The third requires the attacker to guess two groups of code letters and all permutations of where they might be located, which is a few more orders of magnitude harder.
