System Information
OS: 64bit Lubuntu, kernel 4.2.0-23
CPU: AMD FX-8350 Octacore
GPU: GeForce GTX 650 Ti 2gb

Blender Version
Broken: 2.76b f337fea , 2.76-dev fea44ed
Worked: ?

Description of error
I have found that when rendering instances of hair-covered objects (in this case, a small park covered in grass), the BVH build time seemingly skyrockets compared to when non-instanced versions of those objects are copied. By "instanced", I am referring to empties set up as dupligroups.

For example, if I render two copies of the small grassy park (neither of them instanced), then I see a BVH build time of roughly 24 seconds in a quick test. But when I go to another scene and render to dupligroup copies of that small grassy park, the BVH build time is increased to a minute and 20 seconds.

During both tests, I was monitoring CPU usage. In the scene with the increased BVH build time, I saw that only one CPU core was being used during BVH build. This means that for some reason, the BVH build process is not taking advantage of multithreading when a scene contains hair-covered instance objects (and perhaps other types of instanced objects).

Exact steps for others to reproduce the error
Based on a (as simple as possible) attached .blend file with minimum amount of steps

1. Download and open the following .blend file:
   ParticleInstance_NoThreadedBVH.blend901 KBDownload
2. Switch the scene to Original and then render. Note the time for BVH build.
3. Switch the scene to Instanced and then render. Note the increased time for BVH build.