At least 20% of the comments left on the Guardian website each month come from only 2,600 user accounts, who together make up just 0.0037% of the Guardian’s declared monthly audience.
— Belam M. (2012) The Guardian publishes stats via Daring Fireball

This is exactly the reason why any gathered feedback must be balanced with other data points, like hard data from analytics.

Imagine for a second that these people are requesting a feature for your app. They are loud. They write a lot and they request that feature. But then you remember the number above, and you say to yourself: yes, they are good users because they take time to give feedback, but is that feedback representative of the other 99.9963% of the users?

That’s the question.

A question that can be answered in other ways, and you should be in the position to have more and differentiated data points in order to validate the feature request. Raw data from the server interpreted correctly, interviews, surveys, and so on.

Now, it’s also important not to be swayed in the opposite direction, because yes, they are very few, but the fact they are vocal means that they are likely vocal also in talking about your product if they are happy about it.
Plus, there’s the strategy you are pursuing. Does that feature align or not?

All these things are very important and must be kept in mind.