Necessary disclaimers can be found here.
Okay, so now you've had a chance to study NSF's award statistics, it's time to think about the scope of an NSF proposal. Let's face it, a five year, four/five specific aim proposal probably isn't going to fly. Most proposals funded by the NSF appear to be for around three years and have a modest budget. Poke around here to see what the NSF has funded in your area recently to get a better idea of what you're dealing with.
The budget is all important here - it will define the scope of what you propose to do. Remember, those median annual award sizes are totals - direct plus indirect costs. Obviously if you are at an institution with a high F&A rate (say 90%), you're going to be able to ask for more than the median amounts. But don't get greedy - the NSF has limited funds. If your F&A rate is more modest (say around 50%), think in terms of the median amount. Figure out what the median direct rate would be - that's roughly what you'll have to work with (in practice, if you're funded, you're likely to take a modest budget cut, but don't worry about that too much). Let's say you're applying to the BIO directorate, MCB organization, Molecular Biophysics section (one of the more generous sections and the one I'm funded through), and have a F&A rate of 50%. The median direct costs (using the 2007 median annual award) are then:
Basically $100k/year for three (maybe four) years in this example (remember - your mileage will vary). So the important question becomes what can you do with that? The answer will define the scope of your proposal.
What you can do with that amount of money is highly dependent upon the kind of work you do and the type of position you have. I'm in a college of medicine where each PI pays the stipend and benefits (but not tuition - yet) of their graduate students (no TA's). I might just be able to pay two students based on an award of $100/k per year if I'm very, very careful (i.e. miserly). Or one postdoc and part of a tech's salary. Etc. These considerations clearly limit the scope of what I can propose to do with an NSF grant. In the end I've found two specific aims are about all I can manage. If you're in a setting where TAships for grad students are plentiful, perhaps you can manage more.
Something to keep in mind while we're on the topic of the budget. The NSF will only allow you to pay two months of your salary per year from all sources of NSF funds you have access to. Recently there was a discussion at DrugMonkey's place about this. Let's take me as an example. I'm PI on an NSF research grant and on an NSF REU Site grant. If I get 0.5 months salary from the REU grant I can only take 1.5 months salary from the research grant. NSF doesn't appear to care how much of your salary comes from grants from other agencies.
One last thing while I'm thinking about it. Should you get an NSF award (congrats!), there are two types of research grant; Standard and Continuing. You have no say in which one you get. If you're awarded a Standard grant you're given all the money up front. If you get a Continuing grant you get one year's money at a time. Each new year is contingent on an annual report you have to submit (you have to put these in with Standard grants too). I've always had a Continuing grant and have never had an issue with obtaining the next year's money (I am quite diligent about putting in the required report). I've also never heard of anyone not getting their next year's money, but I suppose it's a possibility.