Kiersey cites them in his book, which I don't have a copy of right now. They tested couples and their satisfaction with marriage, and then formulated type recommendations. This is also how the career counseling angle works. They actually test people in various careers and then find who is happiest and most successful, and this is what forms the basis of recommendations for career counseling based on MBTI results. It's not subjective but based on what tends to work out well in terms of happiness between people in relationships and between person and career in large populations.

For relationships the results are sorted not by the most common pairings, but the most successful.

As far as INTJ meeting ESFP, it does happen. People meet in bars, get references from others, meet in church. Perhaps INTJ is tutoring math to ESFP who is studying to be an elementary school teacher and they start dating.

In a relationship this is a pretty good entrepreneurial husband/wife pairing as well. Husband makes the product, wife does sales and marketing (ESFP specialty) and finds out from customers what they really need. ESFP in this case will help him understand and interface with the bureaucratic mentality, and also provide a front face and extremely enthusiastic explanations of his incredible value to the company he works for, or the investors who are funding him, or whatever he is up to. She'll also naturally bring him into contact with these investors, partners, or mentors and massively leverage his life opportunities, which will lead to a lot of success and happiness.

The world is not made for INTJ or INTP and does not understand either of them. The right partner can and will completely solve this problem.

With opposite pairings, there is no friction or stress from one trying to change the other because they are so different that sort of change is thought impossible, it would be like an elephant trying to change a giraffe into an elephant.

Socionics, which is a Jungian typology that's similar to but different from MBTI developed primarily in Russia, provides some of the evidence for this. The types in Socionics don't map one-to-one onto MBTI types, but the general idea of what the grandparent is saying is supported by the concept of "dual" types within Socionics. The basic idea behind a duals is that you get along best with people who share your attitudes/proclivities, but have complementary strengths/weaknesses within those dispositions. And yes, meeting duals is often pretty hard, especially for certain pairs, because they move in different social circles and environments.

It's very interesting stuff and it seems to me like it's onto something, but it's still pretty fringe, and has not been convincingly statistically validated yet. I'm very interested in formalizing these notions and subjecting them to statistically well-founded empirical tests. Right now there's a lot of enthusiasm for Socionics in Russia, but not much work in the Western world. Email or IM me if interested in chatting about this.

Also, since it's so new, beware of crackpottery... there's some people who claim absurd things like being able to determine someone's dominant Socionic type just by seeing a photograph.

OK, finally found a list of studies that discuss complementary type matching.

http://www.capt.org/research/couple-mbti.htm

Liberty's dissertation on attitude-antagonistic coupling covers his research with couples who have the worst-scenario matching, an example of which would be INFJ and ENFP getting together. These pairings will generally make each other completely miserable.