As Razib and I were discussing, the four Bnei Menashe Jewish samples from Behar et al didn't look right since Bnei Menashe are from Mizoram in the northeast of India and thus should be expected to have some East Asian admixture.
When I tried to confirm the admixture/PCA results for Bnei Menashe in the Behar et al paper, I didn't find any mention of the group. Instead, the South Asian Jewish group they mentioned was Bene Israel. According to their admixture and PCA results, Bene Israel looked more like Pakistani populations than their Indian host populations. This is consistent with what my admixture runs show.
So I suspected that the four Bene Israel samples mentioned in the Behar et al paper were accidently labeled as Bnei Menashe in the dataset. I sent an email to the authors and they have confirmed that this was the case.
I have corrected all my spreadsheets so you should see Bene Israel instead of Bnei Menashe now. If you spot Bnei Menashe anywhere, please let me know.
PS. Also, it has been confirmed that three Paniya samples were mislabeled when the data was submitted to the GEO database. They are working on fixing it soon.
UPDATE: Mait Metspalu tells me that the database has been updated with the fixed version of the Behar et al dataset.