Right, so we can actually estimate this, if we have some faith in Zipf's Law as a model for character popularity, and assume that popularity will roughly follow total appearances.

The a and b parameters in 1/(rank+b)^a can't be entirely arbitrary, since the sum for the first 9 ranks—representing the main cast, each credited in over 100 episodes—should be plausible (probably over 90% of people's favourites). So adjustments to one necessitate changes to the other parameter. Playing around this way, we can look at the popularity of the 23rd most frequently appearing character, Kai/Vedek Winn (14 episodes).

Even assuming a lower rank, since the global audience for DS9 is at least 10 million, you need quite large values for a to create a model where not even one viewer regards Kai Winn as their favourite. Much lower ranks (small changes don't make much difference) are needed to change that.

Of course, whether anyone would actually admit to Winn being their favourite is another factor entirely.