I think one should be careful about such rules, while they are useful as rules of thumb I think that truth is that the complexity of code is directly related to the complexity of the domain it is operating within. Where the domain is "messy" the code will be as well.
Simply because a routine has many parameters does not mean it necessarily poorly written or poorly factored. On the contrary it may mean that the code has been well factored and the core routine needs to provide special behaviour in many different scenarios. I guess a clue that it may be such a case is if you find that the routine with 20 arguments is normally accessed only via wrappers which take only a few parameters.