We've given the matter some thought, and this idea most certainly will not be implemented.
What we are proposing is assigning a weight to the problem itself (based on its difficulty, expressed e.g. as the inverse of the number of users who've solved it), rather than to a particular user's approach to the problem.
On the whole the scheme is resistant to dirty accounts. The only possible hazard seems to be a situation of this sort: persons A and B know how to solve problem X, person C does not. Persons B and C are neck'n'neck in the ranks. So, if person A wishes to help person C, he/she creates a load of new accounts and uses them to decrease the difficulty rating of problem X, thus breaking the tie in favor of person C.
Another suggestion that might be implemented one day is that only users who provide some fairly tustworthy form of identification (e.g. a postal address, etc.) will be included in the ranks.
Finally, if the site is to become a little more contest-oriented, we will have to draw a clear dividing line between the user's rank in the online judge, and the user's rank based on overall activity and successes in all official contests. In due course, once these ideas ripen a little and when we can provide clear, sensible formulae for ranks, they will be put under public discussion.