Lightmass appears to bake diffuse lighting data into metallic surfaces' specular response

Lightmass appears to bake diffuse lighting data for metallic materials and then adds it to the specular response of those materials. Is there any way to prevent this? It blows out and reduces the quality of the materials in my mind - I have a reflection environment with a number of probes set up that are being swamped by the diffuse lighting on the metal surfaces.