That never was the goal. The system was created for first open movie project : Elephant dreams.
So, basically, Blender had to respond to the needs of the movie.
At that period, blender foundation did not have manpower to reach heaps of triangles for sequences of movie. And movie plot was made to avoid that.
What was possible was light linking of some properties of lamps.
Shadowing of objects or illumination from some lamps could be limited to objects sharing same layer.
But that was Blender Internal features. Gain was not made on polycount but on shadow quality.
Replacing a distant light source producing costly ray shadows by several close spots doing fast shadow mapping.
Except that, one goal of layers is to have a more manageable scene, blend file.
And the other one, is to have a more manageable compositing in combination of use of renderlayers.
If you only separate objects on several layers, you can hide some layers and work on only one.
But if you render them all at once, there is no reason to expect a benefit compared to a render of all objects put on same layer.
But if your scene can be split into a foreground, a middle ground and a background, you can create a renderlayer corresponding to each ground ; and composite them into an image that could be impossible to render as a unique renderlayer.
For example, it is not rare to render an ocean as separated objects.
Highest picks of waves of ocean object at foreground being able to mask boundary of ocean object of background masking horizon of sky texture.
But if you try to render a unique Ocean object of several km with a close-up on one wave, you may not have enough power to create such mesh.
So, for renderlayers the answer is : yes , that helps.
But for layers, the answer is : we did not really care and it will no longer matter. In 2.8, layers are replaced by collections.