The two big ones are dictionary and brute force. As you know brute force is pretty inefficient when the number of character becomes large. Sure people could develop a statistical analysis for password cracking, but the truth of the matter is that no one really cares. Most passwords are not harvested from cracking passwords, they are socially engineered out. Those that are not social engineered are cracked from either vulnerability in code, or taking from a database that probably didn't salt the hashes.
For example in the news these days people are dumping password files in the MB range, none of those passwords were brute forced or cracked.
Now if we look at common password cracking tools out there, for example John the Ripper (my personal favorite), you see that it doesn't really employ any 'smart' techniques. You can brute force, you can use a dictionary (or other word list), or you can use top common passwords.
In the future I could see password crackers becoming personalized (like medicine), you don't want to have a huge word list of passwords a target would never think of, you want to target that person and you know things about this person. You could give your tool the name of a social networking site and have it mine data, abstracting passwords. Or you can give it other personal information.
One thing to note is, it is always easier to attack the person than the machine (generally). Thats my two cents on this.