There certainly are some computer scientists who are multi-disciplinarians. At some (but not all) of the top departments, there's also a lot of collaborative work that is done among specialists in different areas.
But I think the tendency to specialize on one or two subfields is pretty understandable. A major part of becoming a successful academic is establishing a reputation among the members of your subfield: regularly publishing at the prominent conferences and getting your name recognized as someone doing good work is a major component in whether you get funding, tenure, awards, and the like. It is easier to establish that reputation when you concentrate on a single subfield and a single community of academics. Trying to become prominent in a bunch of subfields simultaneously is not impossible, but it's a lot more difficult.
It's also a lot more effort to do good work in several unrelated fields (and especially to do good work as both a theoretician and a practitioner). In each subfield, there's usually a vast literature that you need to master; learning n subfields is obviously harder than learning 1.
But I think the tendency to specialize on one or two subfields is pretty understandable. A major part of becoming a successful academic is establishing a reputation among the members of your subfield: regularly publishing at the prominent conferences and getting your name recognized as someone doing good work is a major component in whether you get funding, tenure, awards, and the like. It is easier to establish that reputation when you concentrate on a single subfield and a single community of academics. Trying to become prominent in a bunch of subfields simultaneously is not impossible, but it's a lot more difficult.
It's also a lot more effort to do good work in several unrelated fields (and especially to do good work as both a theoretician and a practitioner). In each subfield, there's usually a vast literature that you need to master; learning n subfields is obviously harder than learning 1.