Is a benefit of using joint genotyping in reference confidence mode a more accurate genotype for samples, especially non W Eurasians (when using NA 12878 as the reference) ?
Please see my answer to your other question at http://gatkforums.broadinstitute.org/gatk/discussion/9941/reference-for-wgs#latest.
The benefit of using joint genotyping in reference confidence mode, yes is the power that the cohort provides and also the minimization of compute needed when running per-sample analyses.
Sorry meant hg19 reference, not NA 12878. So I am guessing that the accuracy for variant calls increases because a cohort compromised of like samples from the same geographical area lessens the effect of the hg19 SNP on the called variants, which is the desired outcome.
Yes, I believe if your cohort is comprised of folks that share more recent ancestors, then you can empower calling with confidence variants that may be rare in the overall population but perhaps not so rare in your cohort. An extreme and commonly used example of such an empowered approach is the use of trios--mother, father child.