In general, the reason for baking a component is to carefully remove all the moisture from the plastic part of the component. When a SMT component goes through a reflow oven, the temperature of the component (obviously) rises very quickly, causing any moisture inside to turn into steam. This expansion of water vapor can crack the component, resulting in an unusable or crippled board.
Some components are more sensitive to moisture absorption than others. Once components have absorbed too much moisture, it is a very tedious process to remove the moisture, usually requiring 24 hours or more in a special baking machine. Some of these machines bake the parts in a vacuum chamber.
 

Usually there is a cool-down ramp that is specified by the specific JEDEC standard relating to the solder composition and the part recommendations (in the case of reflow). Most reflow ovens are designed to allow this ramp to be configured (i.e. -5C/second) and then follow that pattern, cooling down by no more than that amount in a given interval. For baking-out moisture from the package, the logic presumably remains the same.
This is intended to remove the possibility of thermal shocking the component or upsetting the solder characteristics by changing the temperature too fast (components are made of different materials, these different materials having different expansion coefficients). The old chemistry lab example of taking a hot Pyrex flask and setting it on the cool lab counter cracking it is an example of what the cool-down ramp intends to avoid.
So in short, if the oven does follow a cool-down ramp, you're good to go as soon as it finishes. If it doesn't ramp the samples down, I'd allow them to come down significantly in temperature naturally before removing them.
JEDEC standard J-STD-020D.