At the micro level, the problem is uncertainty. The "true" or "fair" price of a security isn't something you can just calculate: it's uncertain and subject to change when new information is introduced into the system. So theoretically, when the price gets very far from the fair value, market makers should become more and more eager to trade, since they'll make a large profit when the price comes back. But in practice, they never know whether the person on the other side of the trade is just panic-selling and moving the price farther from equilibrium, or if they know something the market maker doesn't and are moving the price closer to a new equilibrium. As the price gets farther from its previous stable value, the probability that the counterparty is informed grows, and eventually no market maker is willing to trade and the market dissolves.
This failure mode is related to the "Market for Lemons" problem: asymmetric information causing a market breakdown (http://en.wikipedia.org/wiki/The_Market_for_Lemons).
God no. How many stocks even pay dividends any more? The ones that don't are, strictly, "worthless", so their global optimum price should be zero (I'm simplifying; if they come with voting rights then there's a non-zero value to that, but you get the point). It's a bunch of baseball cards at a swap meet.
The disconnect here is that there are agent effects. People make money selling and buying them, so they are incented to push the idea that selling and buying them is a good thing.