The presence of those groups was creating some weird effects in admixture runs at K=8,9. Basically, the ancestral components for Africans I was getting were not stable. Instead they were varying with/without different Harappa participant batches. Also, at K=10,11, there were too many Africa-only ancestral components, forcing me to run even higher values of K.
Since we are not really interested in African diversity in this project and any African admixture among South Asians is most likely to be East, West or North African instead of Pygmy or San, the removal of these groups should not have any implications for the Harappa Ancestry Project.
To make sure that the above assertion is true, I'll re-run admixture analysis for K=2-5 and update later with the results.