I've recently run into an interesting issue, scalability doesn't necessarily go both ways.
Now, intellectually I understand that you make trade-offs to scale to millions of users but that doesn't help my "gut feeling" that I am doing something wrong when I accept the trade-off.
Let's take the classic example of triggers. Triggers tend to be very wrong when it comes to performance, however they are highly effective at securing the data and ensuring business rules are followed. I weigh the need for triggers very carefully before I add them to any table. This can be much more effective than adding the logic to every system that must integrate with the database.
Small databases allow you to pull all sorts of fun tricks that you would never consider otherwise.
Performance concepts do not always scale down to provide the flexibility that is needed on some projects. I've been measuring system with 500GB as a baseline for a while now and forgetting about the ~100GB options.
Right now, I'm looking at different ways to do row, column and cell level security. Almost none of these are viable in a large database. In fact, most of these are not even viable in a "small" database but this gives me a lot of freedom to experiment with concepts, which is fun.