First of all, this is all hypothetical; I'm not actually doing this in a project. For example, let's say I'm trying to make a shape that can either be made by modeling it by hand, or I can get the same shape by applying a subsurf modifier to a cube. Which renders faster, or does it even matter since in the end they are both exactly the same shape?
While it is true that an object with a high number of polygons might indeed take the same time to render as another with a smaller polygon count, but using a subsurf modifier yielding the same number as the first one, but using the subsurf modifier will give you more flexibility and a more responsive interface
Having said that, you do have to be careful with the amount of subdivisions set by the subsurf modifier because it will definitely affect render times and the amount of memory used.
You will probably not notice it on simple geometry, like the example you provide, but once you get into more complex geometry, if you increase the number of subdivisions in the modifier, you can easily push your computer to its limits and potentially make it crash.
The subsurf modifier is a great tool to easily create and edit complex shapes and still have a responsive interface, while keeping a non destructive or permanent workflow.
Creating virtual subdivisions on your geometry saves you the trouble of creating smooth curved surfaces, but when you do the final render, blender has to compute all of that "virtual" geometry as if it was real anyway.
Please read a more detailed post here: Blender render crash when using subsurf modifiers.