The EMH says that you cannot, in the long run, make money on your prediction that there will be a 20 % rally tomorrow, because this, along with the relevant probabilities, are already priced in.
This is related to something that used to confuse me too: the most thickly traded commodity futures are extremely efficiently priced. Yet they exhibit clear seasonal patterns. Why wouldn't someone just buy in the low season and sell high for a near-guaranteed profit? The keyword is "near" -- the prices are such that they counterbalance the risk of deviation from the seasonal pattern.
To answer your specific question about what happens when markets drop: information happens. Events can have nth-order consequences that echo through the markets for months or years, as we find out more about them.
Edit: I should also say that market efficiency isn't a black and white thing. A market can be efficient to me even if someone like Ed Thorp can find mispriced assets.
The market is only completely efficient in the limit. For every mispricing someone finds and exploits the market gets a little efficienter.