The number of people tested by the Geographic Project is mostly Y-DNA and mtDNA tests from version 1.0, which didn't include any autosomal DNA tests. Ancestry probably does have more people in their autosomal database.

However, neither one uses the customers who've tested to calculate BGA statistics. If they did, it would be completely worthless, as their ancestry is self-reported, hence very unreliable. We still don't know what reference populations Ancestry is using, but National Geographic does probably have access to more. Spencer Wells has been collecting samples from indigenous peoples all over the world, plus they could use the same academic databases 23andMe, FTDNA, and probably Ancestry already use.