To be honest, though, I would expect Dorian to still do better than Iron Bull or Varric once they get in the "real" game statistics. The BSN is way more into non-humans than the general numbers turn in. Whatever they get here, you can knock off at least a few points and tack on a few more to humans. I would bet on that.
To a degree, if you're looking purely at "percentage of total players", romanceable female characters will always be the most popular. Their numbers will skew higher, whether they're gay or straight (as in "were they romanced by a PC of the same gender or not"). Numbers for white, blond, and/or "typical" (I don't know a phrase that adequately encompasses concepts like "girl next door" or "handsome jock") romances will also skew higher. Does that indicate racism, or simply an attraction my the majority of players to characters that reflect themselves? I'm not sure you can cite racism when it comes to sexual attraction, as there's absolutely no metric that dictates what someone "should" be attracted to.
Regardless, other than the complication of characters who aren't always recruited (like Fenris or Isabela in DA2), the percentage use of content is pretty predictable when it comes to this stuff, when one is looking at numbers of a certain statistical size (which our telemetry would be, numbering in the millions of players).
Fortunately, we don't write romances to be popular. One need only look at our lineup for DAI to see that's true. We are perfectly comfortable with the idea that someone may play the game either not desiring any romance (which is, in fact, the majority of players--another factoid that does not align with hardcore fan norms) or finding nobody that's personally appealing. We're not going to go out of our way to make a character a niche interest, but neither are we collecting data to distill the "perfect mass appeal" archetype... as I think we all know what that would look like without needing telemetry data.
And, since some people have asked:
* Yes, telemetry does take into account players who do multiple playthroughs. We have data on how many people play through the game more than once, how many start the game multiple times, and the behavior of those who do multiple playthroughs versus those who only play the game once. We know how far those play-once people get in the game before they typically begin dropping out, etc.
* The use of that data is limited. We do not know why players chose these things, for instance, and we have to be careful when we use the data as justification for something. Also, the existence of a choice has inherent value regardless of the number of people that chose it... if that wasn't true, we wouldn't provide evil choices in games (as 86% of players consistently make the "good" choice every time one's presented). At best, it provides us reasons to look at ways to better present/write under-utilized options, or do more work on the things we know will be more widely seen.
* We're not likely to ever provide the data publicly. Maybe we would--but what would be the point, other than to give the impression amongst a lot of ill-informed people as to what we use that data for or what that data actually represents. For every person who assures us this wouldn't cause any issues, we have historical data that says otherwise (see "you can tell us stuff, and say that it's very subject to change and so forth... we'll understand, and not treat it as an ironclad promise").