@Kadri, your results agree with my expectation that an imported OBJ without displacement renders much faster than an object with dynamic micropolygon displacement. One the big expenses in rendered displaced surfaces is calculating ray intersections with micropolygons (for displacement), and this is more expensive whenever you have the unfortunate combination of (1) the rays travelling widely varying directions as they do when reflected off a rocky surface and (2) low roughness setting on the reflective shader, which doesn't let the renderer make any approximations that could make it faster. When you use an imported OBJ with the displacement "baked in", the ray intersector simply intersects static polygons, and this is much, much faster.
However, if you have displacement shaders applied to the imported OBJ (which will be rendered as bump mapping), they can still contribute to render time because they are evaluated when rays hit them. If there are many expensive displacement (bump) shaders applied, it could turn out that the OBJ takes longer to render than the original displaced micropolygons. That's because the micropolygons cache the results of the displacement shaders (up to the capacity of the Subdiv Cache) - i.e. the micropolygons are displaced already - and the displacement shaders aren't evaluated each time the micropolygon is hit by a ray.