In All-I the compression is done in each frame is restricted to that one frame. In IPB, the compression is done across frames, by taking advantage of parts of the image that do not change from frame to frame.
I think the only real advantage is that you can cut to whatever frame you want with All-I. With IPB, since any given frame might be reliant on information in a previous frame, you are somewhat restricted in this capacity. However since you are being less restrictive on the compression algorithm (with IPB), you might get more detail and less macro blocking.
Does this make sense?
I think IPB should be the default choice.