I've wondered about doing just this; putting code in a database to be executed by a parent process, but haven't gotten around to tinkering with it. I even envisioned the concept of pure-Perl modules residing within a database instead of the filestructure. I honestly haven't given it too much thought yet, and haven't explored the issues.

But I do wonder why it's a "bad idea".

You lose all of the benefits like having an editor, having the suite of proven tools like cvs and grep, and being easily able to move the site to another machine and to test out/factor out single parts without having the whole framework in place.

On the other hand, you gain not having several small snippets of code lying around, all dissimilar, you have no issues with two people committing a change simultaneously, and if you put the time into it, you even get a decent editor and the code searching tools you are accustomed to.

If you put the time into it, you can even write a nice and tidy regression suite with lots of metadata attached to each test, but you will always have to have the full framework up and running before you can tinker with any part of the system...