When using Single-Pass-Stereo rendering (I'm using an HTC Vive), this building model renders double vision when more than one instance of it is in the scene. I think it renders correctly when only one is in the scene, so maybe that is a clue to help solve the problem (instancing maybe?).

2. How we can reproduce it using the example you attached

Try running the scene and note that the building renders in an incorrect position causing it to draw double vision. Note that I tried the workaround (recommended by Unity) of importing the built in shaders.

Repro'd here:

Version 5.4.0b13 (90500643c620)
Wed, 30 Mar 2016 04:43:21 GMT
Branch: 5.4/staging