(I think the example I was reading about when I had that thought was navigation: supposedly, men navigate using a mental map, while women use landmarks. "Well," I thought, "I navigate by playing a mental video of myself walking the route I'm supposed to take." Landmarks are part of it, but equally important is remembering turns and orientation. And both of those approaches are a lot more abstract than mine --- the archetypal man constructs a mental map he works from, and the archetypal woman remembers a list of landmarks. I record a complete visual memory of myself walking the entire route, and play bits of it back at will. It's less efficient, but more comprehensive.)
Over the course of the long article I linked to in the first paragraph, Baron-Cohen develops a concept of the male brain autistics are supposed to exemplify: it excels at mathematical reasoning, navigation and spatial reasoning, spotting single elements within larger visual wholes (e.g., the Embedded Figures Task), rotating objects, and aiming and targeting projectiles, and does poorly at verbal tasks, inferring another person's state of mind, cooperation and social judgment. It is also more strongly lateralized than the female brain, with a smaller corpus callosum and more exaggerated handedness. The female brain, by contrast, specializes in language, empathy and cognitive dexterity (e.g., rapidly coming up with a list of objects that are blue, or a list of words beginning with a given letter, or quickly categorizing or matching objects).
As Baron-Cohen suggests, this theory would quite neatly wrap up the past decade's major observations about autistic cognition: deficits* in theory of mind, central coherence and executive function. As defined above, Baron-Cohen's "male" brain type encompasses all three of these: poor theory of mind is practically written into the definition of the male brain (it is, after all, defined in opposition to the highly empathic female brain), so that one's a freebie; weak central coherence is evident in the male brain's focus on detail at the expense of broader patterns, and you would expect executive function to be weaker in someone with a strongly lateralized, weakly interconnected brain than in someone whose brain is more unified.
Unless Baron-Cohen can prove that all these variables (sex, language ability, empathy, brain lateralization, autism etc.) are directly related, all his theory will have accomplished will be redefining normal maleness as a state of mild autism. So far, the evidence is inconclusive. Yes, the few male-female differences in cognitive style that exist (male advantage in mentally rotating 3-D objects, female advantage in verbal fluency) are consistent with Baron-Cohen's ideas of the male and female brains, but other areas he predicted would show differences haven't, or have shown them inconsistently (empathy, mathematical and language ability, interest in social vs. mechanical stimuli) --- and even the differences that have been shown to exist are small ones, with a substantial zone of overlap between the sexes.
Language ability is also quite readily shown to be independent of social understanding by the example of Asperger syndrome. Baron-Cohen conflates autism with Asperger syndrome in his article, and speaks of both as involving a delay or impairment in language acquisition and use. This isn't true; one of the diagnostic criteria for Asperger syndrome is that language acquisition must occur on time or precociously. Baron-Cohen places AS on a continuum between the normal male brain and the "classically autistic" brain, which is problematic given AS's normal or superior verbal capacities and its higher male-to-female ratio (9:1, as opposed to 4:1 in classical autism).
Neuroanatomically, the theory is on ground as shaky as it is in cognitive psychology. Baron-Cohen invokes differences in the size of the corpus callosum to explain the contrast between the strongly lateralized male brain and the adapatable, bilateralized female brain, ascribing the female advantage in verbal fluency to speedy communication between hemispheres. Unfortunately, this 1997 meta-analysis of 49 studies and this current study indicate that there is no difference between male and female corpora callosa that cannot be accounted for by overall differences in brain size --- which favor men. Also, autistic people (who are definitely less fluent overall in their use of language than NTs) tend to display very weak brain lateralization and handedness, which Baron-Cohen acknowledges.
I can understand the appeal of a theory relating male gender and autism --- it would explain both the preponderance of men on the spectrum and the trouble many female autistics have fitting into the feminine social role. I think it would ultimately obscure more than it illuminated, though, both about autism and about the nature of gender. Defining maleness as mild autism would enshrine in biology a very culturally specific ideal of geeky masculinity, and legitimize (again, via biology) the delegation of all familial and relationship responsibilities to women, who, after all, have specialized nervous systems that can handle all that stuff.
*Autism research tends to find "deficits" in autistic cognition wherever it goes looking for them, even when the actual results indicate an autistic advantage --- e.g., findings that we tend to do better on visual-search tasks getting interpreted as showing "defects in central coherence," "can't see the forest for the trees."