I find it makes more sense to have most of the ‘functional’ parts of your code either in modules outside the scene file, or in functions separate from the scene methods: create(), show(), etc.
Why? Because, all too often, I see users (new and old) write long, long, messy, long, complex, … you get the idea, code in the scene methods, and then because the file is so long, they goof up basic stuff about scene management. i.e. It just gets to be a disorganized mess mixed up with the scene management code.
However, if you put all the code that does building and other work in a separate module, you can simply require the module in a lightweight scene file, pass the scene group to the module (for managing objects created there), and keep the scene file easily read and managed. This also makes your code much more portable in case you want to re-use bits of it in the future.
One thing I didn’t mention though, is project size. If you’re doing a very short or small project (less than say 1000 lines of code), you might be fine embedding all the functional code in the scene management methods. Adding a module structure to a small project may simply complicate things for no purpose.
Here are some samples of composer scenes I put together for the Corona Geek Show (dig through the git for more):