Massive sampling bias here... These are passwords from sites that have been hacked... So only tells you that people use crap passwords for sites that can't be trusted.
Also, these passwords were either A. stored in plaintext, in which case complexity is irrelevant, or B. stored hashed, in which case only easily crackable passwords would be released; skewing the results even further.
ALSO... as all the 'good' passwords are probably unique, they will never be at the top of the list of passwords. So the non-unique passwords will inherently have more people using them.
The more I think about it, the more meaningless this information becomes.