Ongezellig

Kinship coefficient (F) describes how more probable it is that a related person has the same gene as you when compared to an unrelated person. By definition (p_y = F_{xy} + (1 - F_{xy})p), where (p_y) - probabilty that the person y has a given gene, (p) - frequency of that gene in general population. Fixation index (F_{ST}) describes the same thing, but between subpopulations, so if you take a person from the same subpopulation as you, (p_{same} = F_{ST} + (1 - F_{ST})p) and if you take a person from a different subpopulation, (p_{other} = - F_{ST} + (1 + F_{ST})p) (at least in case of subpopulations of identical size, we are doing the simplest case here), which means that your kinship with a person from the same subpopulation is (F_{ST}) and with a person from from different subpopulation is (-F_{ST}).
<Maya space
In an uniform population (not divided into subpopulations), the kinship between a person and his child is 1/4. Choose a locus and a gene at that locus in yourself, there is 1/2 chance the child inherited this gene from you (as opposed to inheriting the allele from the other chromosome), and there is 1/2 chance you have picked that gene in your child (as opposed to picking a gene that it inherited from its mother.)
<BWC space
If we take subpopulations into account and if the mother is from the same subpopulation as the father, the kinship with the child jumps to (1/4 + 3 F_{ST}/4). The reasoning is as follows: with probability 1/2 you pick the allele it inherited from you, then with probability 1/2 is the same allele you've picked initially and with probability 1/2 it's the other allele that comes from you, which in turn is identical to the first allele with probability (p_{same} = F_{ST} + (1 - F_{ST})p). And with probability 1/2 you pick an allele coming from the mother, who is from the same subpopulation as you, so the probability of it being identical is also (p_{same} = F_{ST} + (1 - F_{ST})p). Finally, you have (p_{child} = 1/4 + 3/4 (F_{ST} + (1 - F_{ST})p)). Kinship coefficient is equal to (F_{xy} = (p_y - p)/ (1 - p)) after transforming the first equation. Transforming (p_{child}) gives:
• (p_{child} = 1/4 + 3/4F_{ST} + 3/4p - 3/4F_{ST}p)
• (p_{child} - p = 1/4 + 3/4F_{ST} - 1/4p - 3/4F_{ST}p)
• (p_{child} - p = (1 - p)(1/4 + 3/4F_{ST}))
• (1/4 + 3/4F_{ST} = (p_{child} - p)/(1-p))
The above means that (F_{child} = 1/4 + 3/4F_{ST}). In case of a mixed child (F_{mixed} = 1/4 - F_{ST}/4) (I'm NOT rewriting the derivation.)
<mutt space
Now the question is: at what (F_{ST}) the mixed child will be as related to its father as an unrelated person from his own subpopulation, i.e., when (F_{ST} = 1/4 - 1/4F_{ST}). That's a simple equation, the result of which is (0.20), which means that if a person chooses a partner from a population that is more distant than (0.20), his child will be less related to him than for example an adopted child of his race.