Last year I counted the three hours as follows:
Brainstorming (I laid out three ideas for games, basic plot and PC tasks/interactions) and ran them by a couple friends. This was to choose which game(s) I would make. I didn't count this.
Planning 0-0:15 - This is where I diagrammed
Pete's Punkin Junkinator. The map, a list of characters and objects, and a list of requirements for completion (winning).
Execution 0:15-2:15 - I pumped it out.
Testing and corrections 2:15-3:00 Part of this time was originally slotted for
Spiral Cheese Horizon's music composition, but it was scrapped for many reasons. I played through a few times and added to the game AS I WENT. The debugger is your best friend in this speed testing because you can simulate your corrections in game without having to restart it. Make sure to test at the end!
I don't remember if I wrote a walkthrough, if so, it wasn't made in the three hours. I know in the last few minutes I stopped adding details and just made sure the game's win-condition tasks were all completable - which included removing one or two of the originally planned completion tasks. It was missing a lot of features: hints about task completion, a good scoring system, conversation topics, etc. Some things I learned for using the three hour limit:
While my scoring system might have been frustrating for high-score hunters (the max was unreachable), the fact that you only needed to complete 4 tasks out of a set of 7 or so meant that when the clock was close to finished, I could freely decided which to scrap and which to finish. Originally, the design had about 10 win tasks, but since the game didn't care which 4 you did, if one became too hard/time-consuming to make, I scrapped it.
The format for the plot development is one I've used before (in Give Me Your Lunch Money) Cutscene intro, cutscene ending, with filler in between. This format seems to occur naturally in my head but it's really useful for a short timeframe because you can build the skeleton of the game and fill in the meat after. How much meat? How much do you have time for?
If you work with a set of optional tasks rather than an exacting time/plot-line, you can also estimate the amount of time it will take to do by extrapolating the one task, and if you're lucky, you can even copy and paste assets.
It's specific to my game, which was made easier by being as gimmicky as the rest of mine, but there was a lot of talk about how to make the game in 3 hours and I found it to be an inspiring and freeing limitation. Perhaps if someone forced me to use no more than 3 hours fixing it, I'd re-release Mangiasaur!
PS I will be submitting a game for Ectocomp, tonight or tomorrow.
PPS @ Mike - I'm afraid to use ADRIFT 5 because I don't KNOW it yet. If I have to study it for a couple hours before working on the game I will, but I wouldn't count that towards 3 hours. You mentioned a walk cycle you were having trouble with. My advice is play around with another file first or save it for another project. The best way to know how long it will take you to make is to already know exactly how to implement it. If you can't code it into PC interactions but need it in the game, maybe you can make it into a cut-scene!