Research exists on craniopagus twins, maybe most notably Tatiana and Krista, who seem to share sensory input somewhat. I doubt that connective mechanisms such as this abnormal case would suffice to permit "compound cognition" in ways that would enhance cognitive ability similarly to your point about hominid evolution. Your relatively simple proposal for a connective cord of neurons strikes me as roughly equivalent, though more complex connections may not be entirely unfeasible in the long run.
Before I move on to speculating about that, consider a few other points from cognitive neuroscience thus far. Plenty of split-brain research (on people and animals with dysfunctional corpus callosums) exists as well, little of which shows what I would call "radical effects on cognitive abilities". Results are somewhat more subtle than that: e.g., involving object recognition more than what laypeople would recognize as intelligence, especially social communication and basic reasoning, which remain relatively intact. You might also be surprised at the modest effects of losing an entire cerebral hemisphere: again, it's not what I would call a 50% reduction in cognitive ability, at least as we understand and use such abilities in modern social contexts.
Non-biological methods are already being experimented with for brain-to-computer and even human brain-to-brain interfacing (Armstrong & Ma, 2013), but we're some distance away from really augmenting cognitive abilities through such methods in the way evolution seems to have augmented them in hominids. Consider also that evolution seems to have proceeded in far more complex manners than those by which we're connecting brains now. Changes occurring organically throughout the brain over millions of years of natural selection may have developed our cognitive abilities well beyond those of smaller-brained animals, but these changes have probably been much more sophisticated than simply wiring disparate parts together, let alone only one part to only one other. Cognitive processes are functionally dispersed broadly across our incredibly complex neural structures; we definitely use more than 10% of our brains. There's really no particularly promising hub-of-all-cognitive-ability through which one could connect and expect to provide consciousness-wide augmentation (that I know of), nor would it be as simple as connecting one brain to another.
To throw in a halfhearted computer metaphor, consider how little processing ability is gained simply by connecting two computers by LAN cable. In very many ways, computer processing augmentation has had to proceed much more incrementally, component by component. Multi-core processors and several sticks of RAM per computer are commonplace now, but to truly augment computing performance with these, fundamental architectural changes in connective design have necessarily occurred. Far more than one wire connects these compound components.
Thinking in this manner might pave the way to future innovations that may someday achieve the effects you have in mind. Transhuman cognitive augmentation research is rapidly advancing in very many directions, some of which may bear satisfying similarities to your idea. They're all well beyond the scope of my knowledge or of this answer though, so I'll just toss some Wikipedia links at you to get things started. Please feel welcome to ask separate questions to follow up on cyberpsychological topics; I find them all very exciting, and would be very eager to explore them further with you and our community!
Armstrong, D., & Ma, M. (2013, August 27). Researcher controls colleague’s motions in 1st human brain-to-brain interface. Washington University: News and Information. Retrieved from http://www.washington.edu/news/2013/08/27/researcher-controls-colleagues-motions-in-1st-human-brain-to-brain-interface/.