In a world where it was indeed perfectly even, lets say a world with a highly regimented system of prejudice where everyone's skin tone is carefully measured and restrictions were then applied across the country.

But if prejudice isn't perfectly even or measuring the same things then you'd expect it to vary.
If one twin grows up in a town where most of the residents still oppose miscegenation while the other ends up in a town with far less prejudice then the difference is going to show up as environmental.

Or if they have the same accent and same skin tone but one grows up in a town where people are more prejudiced about cultural markers like accent and dress while the other grows up in a town where they care more about biological markers the difference would show up as environmental.

Also it's common to run PCA when dealing with samples from different ethnic origins to make it easy to spot when you've simply flagged up a variant linked to one group.

Examples:

http://www.genesandhealth.org/sites/elgh.mrmdev.co.uk/files/styles/380/public/Screen%20Shot%202015-02-26%20at%2010.28.28.png?itok=PyhQvYTn

https://static-content.springer.com/image/art%3A10.1186%2F1753-6561-5-S9-S116/MediaObjects/12919_2011_Article_1167_Fig1_HTML.jpg

If your flagged variant overlaps almost totally with only one group then it can be a sign that your analysis has been confounded though overlapping a little more with one group than with others is to be expected.