Why would they cancel each other out? Why would there be equal numbers?
Why would you assume that? To me it seems the opposite, that they wouldn't be evenly distributed.
In areas like SFBA where the job market is liquid, people will be less likely to be stuck in a job where offer basics like salary don't match their expectations -- so they'll be more likely to weight other factors like culture and mission when considering how happy they are at their current job. In less fluid markets, people can more easily find themselves in a position where they're disappointed with salary but unable to find a better opportunity.
In those situations, you might be more likely to find people who are either satisfied or dissatisfied with culture, mission, growth etc but are also disappointed with salary.
First, even in your version, the "feel goods" (high salaries) and the "complainers" (low salaries) would cancel each other out, and (assuming equal numbers of them) will give us a usable median value.
But in the real world is more complicated than "self-selection" in the basic mode you describe.
Who might "feel good" and who might feel "justified in complaining" will not be so much based on the salary, but more on the person and their personality (which we can assume are evenly distributed among salaries).
In other words, it won't only be actually great salaries that are reported from people "feeling good", nor only low salaries reported from people "justified in complaining". And then you'll also have people that regardless if they consider their salary high or low, they want transparency etc, and report their salaries.
The end median result, given enough samples, would be a useful, if not 100% accurate, representation of the situation.