The problem she described is a bit tougher to solve than simply saying, "don't hire booth babes." "[W]omen who are pretty but not models, dressed in normal street clothes" describes a pretty big percentage of the general population. Under her definition, "If you have 'booth babes' ..., I'm not going to buy your product" is essentially the same as saying "If you have women selling your product, I'm not going to buy it," which isn't exactly an enlightened position to take.
The problem the OP faced was that people at the conference were generally sorting people into the general categories of "technical and worth talking to" or "just trying to sell me something" and would make the Bayesian inference that a random woman they came across was in the latter category.
I can see two solutions here, neither of which is very good:
1) In order to promote gender equality, everyone at a conference should avoid judging people based on their gender. This would be ideal, but it's hard to fight casual statistical inference.
2) The conference could make sure that the people selling stuff at the booths and the people attending the conference have similar demographic profiles, so a random woman you encounter at the conference is as likely to be technical as a random man. In practice, though, this would mean preferential hiring of men for the promotional jobs.
Really, we need an option (3) of "get more women to be developers" but that's hardly news, and not something that can be solved in short order.