If the extended mind literature is onto something, then any two individuals with a sufficiently reliable relationship can be considered part of a single cognitive system, just as cohesive as the intra-cranial faculties that typically define a cognitive agent. For example, married couples reliably co-represent facts and utilize each other to make decisions with an advantage over individuals solving tasks alone (Berg, et.al. 2007). Even if one is skeptical that the mind is truly extended, the benefits of group recall, problem-solving and decision-making remain.
Now suppose that the act of retrieving memories stored in ones spouse's mind can be faster, more reliable and easier than finding the information within one's own mind. This consideration runs counter to work in epistemology on testimony (see Lackey & Sosa, 2006) which suggests that the combined act of memory and perception required for testimony renders it more difficult than acts of mere memory alone. The problem with this approach is that it assumes that complexity signals difficulty.
A priori, it seems that linguistic communication between married couples is more efficient than between less familiar dyads--hence the amusement of "The Newlywed Game". Older married couples have reliable patterns of information processing and behaviour that automates particular interactions. Combining communicative ease with retrieval difficulties of one's own memory; the total effort required to request answers from one's spouse may be simpler, both experientially and computationally, than piecing together one's own fragmented recollections.
How can this be? Well, suppose that running multiple cognitive faculties in parallel can be more efficient than running a single process serially. For example, perceiving and understanding ones spouse's linguistic utterances involves many faculties working habitually and synchronistically. Contrast this with accessing a dubious past perception. The latter may be perplexing and uncertain. Even if the faculties running in parallel take longer than the serial process to produce outputs, those outputs may yield strikingly different epistemic states. Depending on the context of use, the former may have epistemic reliability that the latter do not. For example, our spouse can help us find our keys if we are absent-minded much faster than we can by-ourselves.
Also, each year cognitive tasks that were once viewed as mind-bogglingly difficult are found to utilize fairly simple heuristics. E.g. tracking a baseball requires maintaining the angle of one's head, which is a vastly simpler task than explicitly solving parabolic functions.
It may turn out that testimony between intimate dyads deserves a unique epistemic approach.
 Of course, any task which is solved through embodiment or outsourcing then gets questioned about its cognitive status. Is the mark of the cognitive only those activities involving thought?
Berg, et.al. (2007) Task Control and Cognitive Abilities of Self and Spouse in Collaboration in Middle-Aged and Older Couples. Psychology and Aging. 22(3). 420-427.