I personally think that quicker challenges are better. I knocked up my box in a few hours because i said i'd enter and i forgot about it.
If i had remembered and knew i have another week to do it, i'd have put it off.
I think the best thing to do is trial run a few short ones over a couple of weeks (best to spread it out as people might be busy at different dates) and then trial run a few longer ones. Keeping the themes and outcomes more or less the same difficulty level and see which works out best. Advertising them over on Geek Culture might help. Prizes also work wonders, maybe even something like getting a member(who can actually model/texture) to model a one off object and offer it as a prize to the winner. One off models are better than offering a pack that is for sale because that model will belong to the winner and the winner only, for use in their projects. Its more of an incentive to win aswell.
I always thought how well a few judge members would work rather than a public vote, public votes can sometimes (unconciously) be biased towards whichever user the voter likes more(just saying). Judges would rate the models on categories (Modelling, Texturing, complexity, overall) and then tally up the scores, if scores are even then a vote is cast to decide the final winner.
Just throwing ideas out there but it can't hurt to throw a few into practice just to test them out. I think making this more of an official thing would be better aswell, newsletter coverage if it ever takes off.
