Anatomy of an NSF proposal

My basic rule-of-thumb is:

1-2 pages placing your work in a broader context - what sort of problems do you expect to be solving 10-30 years from now? Why is the problem you're addressing relevant? This section should also have 1-2 paragraphs summarising the proposed research contributions (what will you have done after 3-5 years?) Don't make this too broad or make it sound like you're doing something you're not - you want to convince them that the direction is worth while and will lead to bigger and better things, but you don't want to claim to cure cancer when all you've done is create a better pipette...
You may also want to include 1-2 paragraphs discussing outreach and education, especially if this is a Career award or you have something particularly compelling to say in this area.
2-3 pages for previous work/work you've already completed/existing research. Be very explicit about work that is yours.
5-7 pages for the proposed work. This will probably include background on the problems. Stay away from comments like "we plan to investigate..." without any specifics. For every problem you should have your specific approach to it. And, ideally, this shouldn't be the approach that someone in your field would come up with after they sat down and thought about it for 10 minutes. Also stay away from the thesis-as-hammer grant proposal - I've seen a lot of proposals that take the basic idea from their thesis and just list a set of problems to try it on. You really want something a bit more visionary.
In this section, the breakdown should be:
- 2-3 pages of problems where you have already done some work and have preliminary results. The listed problems are extensions of existing work/address known flaws/application of current techniques to a new, but similar, problem. You should be 90% confident of succeeding in solving these problems. The solutions should be very detailed/explicit. I.e., this is where you demonstrate that you know the problem area well, you know how to solve the problems, and at the very least, you will definitely accomplish these goals.
- 2-4 pages on longer-range problems, ones where the solutions are not as obvious. These problems might have one or more proposed solutions, and will, of necessity, be a bit more vague. This is where you convince the reviewer to fund your proposal.
- 1-2 pages of long-term, we'll get to them if we can, type problems. This is, in some sense, a peak at what your follow-up proposal will have in it...
2 pages on education/broader impact. It's rare to have anything unique here - it's usually "I'll run this course which will attract these minority students for some reason, and will integrate my research with the curriculum". If you can link into programs on your campus that do education for kids/high school, that's a good thing. If you have any industry contacts who are interested in your work, this is the place to mention them (include a letter of support in this case).
(Optional) Results from prior funding (if any) or anything unique you want to say about your self that makes you particularly suited to this research/outreach.
1 page summarising the proposed research and who's going to do it (e.g., graduate students) and how. Make this easy to find so the reviewers can quickly flip to it and not dig this information out of the body of the proposal. Include your integrated education plan.
A conclusion summarising why this research needs to be done, how it will help the community, why you're excited to do this... i.e., end on a high note. This is your chance to remind the reviewer about why they should be so keen on your proposal...

Budget

I don't think I've seen a proposal killed because of an excessive budget (or a too small one), but it will result in reviewers making nasty comments about you to each other. Just remember that anyone who has a substantial role in the project should have commensurate funding. If you lay out a time-line with research problems and people assigned to them, then that should serve as a check on whether or not you have too many/too few people.

NSF will probably cut your budget anyhow.