I'm not sure I entirely understand the constraints, but a couple of things come to mind. First, you can move all of the distracting code to a package, and you can then just load the package (Needs or Get) from the "main" notebook. Second, you could generate your desired output rather than just give them the "raw" notebook. For example, you could find all output cells or all cells with a particular tag and copy them into a separate notebook for distribution. Or output as PDF or other formats. You can check out some of possibilities here: http://reference.wolfram.com/language/guide/LowLevelNotebookProgramming.html). Third, there are some more basic things you can do within a single notebook, like hiding certain cells. Your inline comments could be simple text cells (I know you didn't ask about that, but it would make the notebook look more like a document instead of a big block of code).

I guess my general suggestion is to separate out your computations from what you want to present. Do all of your computations in your messy sandbox area, and then just generate the output that you wnat to actually present. Export that output in whatever format is appropriate.