There's a frustrating lack of theory and firm empirical research behind dissociative identity disorder. Quite apart from whether the condition actually exists (and I feel relatively comfortable claiming that it does, with the caveat that it may be overdiagnosed), there seems to be a certain... vagueness about why people might have DID. "As a response to childhood trauma" (the traumagenic view) or "as a form of false memory" (the iatrogenic view) seem to be the main options, but there's mostly silence on anything more substantial.
Simone Reinders' work is, in this context, one of the most promising research programs in the area. Reinders' lab works on producing neuroimages of patients with DID, with the hope of eventually figuring out what exactly is going on in the brain of a DID-diagnosed patient. Reinders' current working theory is, I believe, that DID is related to deviant amygdalar and hippocampal volumes - of note, perhaps, is that the hippocampus is more frequently found to be implicated in DID than the amygdala. Respectably, she has also commented that given the rarity of rigorous empirical research into DID we should remain cautious about adhering to any particular view - open minds are the order of the day.
I'll be watching the work of the Neuroimaging DID project closely - it's a research project that's been sorely needed, and it should be fascinating to watch its progress.
Reinders, A.A.T.S. (2008) Cross-examining dissociative identity disorder: Neuroimaging and etiology on trial, Neurocase, 14(1), 44-53
Reinders et al., (2006), Psychobiological Characteristics of Dissociative Identity Disorder: A Symptom Provocation Study, Biological Psychiatry, 60, 730-740
Vermetten et al. (2006), Hippocampal and Amygdalar Volumes in Dissociative Identity Disorder, American Journal of Psychiatry, 163, 630-636