I found at least one good reason to do this: if the code you are writing tests whether a module is available (through an eval { require Module;} and defaults to a less-than-optimal-but-acceptable behaviour if the eval returns an error. Then you need to test both cases, with and without the module. A workaround is to wrap the eval in a test controlled by an environment variable, so when testing you can pretend that the module was not here. This is not perfect (the module could be imported by an other part of yur code or by an other module) but it is usually the easiest way to "desinstall" a module.

In this situation I would simply modify my code to allow me to test both situations by only requiring the module if a certain flag was set (or unset). Then I could just flip the value of the flag whenever I wanted to test the other bit of code. Certainly easier than uninstalling/reinstalling modules.