A document from NASA Safety & Mission Assurance suggests the reason is structural:
Thin-skinned Centaur cannot easily support the giant 5.4 meter diameter payload fairing, so the Contraves composite fairing also encloses Centaur.
My first guess, however, was aerodynamics, particularly the Whitcomb area rule: for transonic flight, abrupt changes of cross-sectional area produce lots of drag.
The Centaur upper stage is about 3m in diameter, and the Atlas V first stage is 4.1 in diameter.
The 4xx configurations use a 4.2 meter payload fairing, so the profile is 4.2m-3.05m-3.8m; the upper stage cross section is 53% of the fairing cross section, which is a pretty significant drop.
If the 5xx configurations left the Centaur un-faired, the profile would be 5.4m-3.05m-4.1m; the upper stage cross section would be only 32% of the fairing cross section. Instead, it's faired; there's only one cross-section transition instead of two (on the 501, at least), and the first stage is 58% of the fairing cross section.
There might also be airflow/turbulence issues necking down further behind the 5m issue that would cause structural problems; ISTR something like that related to Orion on Delta IV, but I could be making that up.
This thread on nasaspaceflight.com discusses both structural and aerodynamic possibilities, as well as thermal considerations on the Centaur (though it doesn't explain why the 4m version wouldn't have the same thermal requirements).