Keyframing means that file stores a snapshot of the model for a number of "key" frames. In essence the transformed model vertices are stored for these keyframes.
To get a smooth animation you will interpolate vertices between two keyframes.

While it's much simpler to display animations this way, storing an animation using keyframing isn't very efficient or flexible. Sub-animations for example require the model to be split into parts (like Quake2 models) to enable separate lower- and upper parts. Skeletal animation has the advatage of just storing a set of "bones" (basically just transformation nodes) that are attached to a "skin" (or actual vertices). Animations can be stored separate from models and can also be blended (e.g. running while looking around). Unfortunately not many open formats support skeletal animations.

HTH,
Pat.