Here's my first stab:

The ancestral environment had little if any interracial couples producing kids. (The old rape-the-neighboring-tribe's women thing still didn't involve vastly different races on the scale of chinese/causasian.) Therefore, kin selection pressures did not perceptably benefit from distinguishing between "siblings from interracial parents" vs. "siblings from more closely-related parents" -- all parents were the latter type.

So it doesn't seem like there would be room for a more fine-grained relative-gene-similarity detector to develop: most of the benefit from helping your kin in the ancestral environment is from the fact of that kinship, and further deviations from the simple percentage-genes-shared calculations (due to the relative relatedness of parents you mention) would just be very weak noise.