Design approach: Overloading vs Switch?

In regard to performance and scalability in package design, is it best to:

1. … ‘overload’ function names (letting Mathematica sort out which version to use based on patterns/conditions/tests and the way the system orders definitions)?
2. … or to construct a single function with a Switch[] (or similar command) to direct evaluation?

Mathematica’s expressiveness frequently confuses me with silly (?) issues like this.