You use a chrome ball, alongside a 50% grey ball, to collect information on the light in a scene for a better composite and for lighting a CG object. For instance in my last gig I used them to place lights in my virtual scene, I had positions already for the studio lights used on the shoot but the chrome reference allowed me to identify specular reflections off other objects in the scene, and place lights accordingly.
After you matchmove your scene, take a CG sphere and place it in the scene at the location your real chrome sphere is in-scene. Now place your lights by matching the highlights, essentially each highlight on your sphere gives you information about angle and intensity of light in your scene.
Sample the color of the chrome sphere's highlights (if they aren't fully blown) to get information on the light color. Environment-map your CG sphere and compare it to the chrome sphere to check the reflection maps you will be using.
The grey sphere lets you analyze the softness of the light in the real-world scene, allowing you to match the umbra and penumbra of shadows, and lets you sample the shadow color against a known color (50% gray.) It also acts as a matte surface to allow you to get a feel for specular highlights in two extremes, a hard reflective surface and a matte, porous surface.
Completely separate from using a chrome sphere as a lightprobe for HDRI rendering, although a good lightprobe can help alleviate some of the need to use reference spheres.
So basically the use of in-scene reference spheres is to give the artist a set of known surfaces against which they can analyze and match lighting.
I always try to get them in a take as a reference, particularly when a Final Gathering/HDRI approach just isn't feasable (as is usually the case in a production environment.)
