The $1700 PC was a nice machine. Perhaps a little more heavily focused on gaming than I'd like to see, still none of its parts raised eyebrows. It was a "safe" build, even if a little boring in some ways. If I won it, I'd probably build it as-is, although I'd likely end up selling the GTX680 because my games just don't need that power.
The $1000 machine used a nightmarish case and made a few uncomfortable compromises. If I win it, I will sell the graphics card to buy a SSD, 8GB of RAM, and a considerably weaker graphics card, and build it for my non-gamer wife in her Antec Sonata III.
The $500 PC was a remarkable experiment. At its specific target, it seems to do quite well, but for anyone who isn't playing games 100% of the time, it would be an exercise in frustration. Should I win it, I will probably sell the graphics card and CPU for an I5-2400, get 8GB of RAM, and build it for my non-gamer father using his HD6670 as a substantial upgrade over his E6750.
Similar to what another poster suggested, I'd like to see one of two notable "requirements" in these builds. They must either be something that a real user would actually build, in its entirely (last quarter's $2.6K build was a perfect example, this quarter's $2K build might have made it if the remaining $259 in the budget had been used), OR, it must be an experimental build purely intended to test something legitimate. It might be nothing someone would build, but maybe only because they'd never thought about it. This quarter's $500 build clearly qualifies, and I hope it is remembered as fondly as that quad Crossfire HD4850 build that a few of us have mentioned. With that latter in mind, some ideas might include physically constrained builds (must fit in carry-on luggage, or be suitable for a dorm room) or machines that must achieve certain minimums in a real-world benchmark (not necessarily a game).
And although I do prefer firm budgets, jaquith has a very good point. $25 here, $50 there, adds up (and where do you stop?), but a single carefully justified $25-$50 "bump," representing perhaps another week of waiting and/or saving, seems reasonable. After all, just think what the $500 PC could have been like with a $100 CPU. It might even be appropriate to allow a "do-over" for a part, such as to avoid a fiasco like the Logisys case, or last quarter's defective motherboard. Afterall, an experimental build could become a total waste due to a single botched part.