I don't think this is even strictly true. It seems to depend on the attribute distribution you are using.
For a simple distribution, let's say both green and purple populations have on average X talent with stdev m.
Green has Y diligence with stdev m, and purple has Z diligence with stdev m, where Y >> Z.
Success is measured by (talent+diligence).
I haven't gone through the math, but just picturing it in my head the original assertion doesn't seem true. Something like this:
Just to use the picture, consider top 33% instead of top 5%. The purple (left) distribution has a decent amount in the top 33%, but essentially 0 in the top 1%.