[jacquesm]

23andme is collecting data at a fair rate, there are other companies like them. Pooling the data between all those companies is going to allow you to fill in a bunch of 'blanks'. At some point that will reach critical mass and you can map the remainder.

I'm not good enough at math to give you the percentage of a certain population in order to be able to infer the rest, maybe someone else here can do that.

But given a population size 'n' if you get a random distribution of individuals and you know their genes and you know have a graph of relationships (say through facebook or some other means of tracing links between people) you should be able to make a formula that tells you what kind of 'coverage' you can expect based on how large a sample.