The scenario presented seems very odd. Why would you want to sort 10^7 items that are known to contain only four distinct values? It seems much more likely you would be counting the number of times each value appears, or selecting all of the elements of value X.

I believe the purpose of choosing such an odd scenario is to show that, while you might think that you can beat the generic sort algos with a more domain-specific implementation, you might be wrong, or you might not gain enough performance to make it worth the other pitfalls of using such algos