"Looking-in" simply means that the cameras will need to be angled such that their principal axes cross over at some point in front of the camera (rather than being parallel as is often assumed for stereo work)
The reason for doing this when the cameras are far apart is also simple:
Stereo requires you to match features from one image to the other. Hence the cameras must both have a view of the objects you are interested in. Parallel cameras only overlap when the object is sufficiently far away - by pointing them inwards, you can solve this. (Of course, if your object looks significantly different from the two viewpoints, you still won't be able to do a good stereo match.)