This theory is derived from a wealth of data regarding protein homologies. It shows that a fully functional flagellum could evolve through multiple functional intermediates. Each step involves the modification of only a single protein. For more discussion on the actual proteins involved and their homologues see the link above.
When ID proponents ask what good is half a flagellum, well, ¾ a flagellum is good for dispersion, ½ a flagellum is good for attachment to substrates, ¼ a flagellum is good for regulated protein secretion.
This model based on a wealth of evidence shows that the flagellum could evolve. Since 2003 many experiments have confirmed what initially were only predictions or weak assumptions. Still, many fine details remain to be worked out regarding function, not origin. For example, how does the ATP synthase derived components impart the motion of active transport, or how does the Tol-Pal derived components actually spine the flagellum.
Behe in the Dover trial concedes that Irreducibly Complex systems can evolve.
Lawyer -- "You say, Even if a system is irreducibly complex and thus could not have been produced directly, however, one cannot definitively rule out the possibility of an indirect, circuitous route, right? Behe -- "Yes." Lawyer -- "And by indirect, you mean evolution from a pre-cursor with a different function than the system being studied?" Behe -- "Yes, different function, perhaps different number of parts, and so on."
To be fair, however, Behe claims that at a certain point of complexity an indirect route is impossible. But he gives no evidence to support this caveat.