When photographing buildings from the ground with a shift lens in order to keep the image plane plumb, the effect is like using a longer focal length lens for the top of a building than for the lower floors. Shifts may keep the building face parallel with the sensor/film, but it also expands the image of the upper floors relative to the lower ones. The same effect occurs when applying stretch in post. Stretch applies the same graduated zoom lens effect just like like tilting the easel in the darkroom to make the same correction. (Of course you lose some resolution in the stretched portion of the print.)

A full and drastic correction, as stated earlier, strikes us as artificial. In person, and partly because our retinas are curved, we do not see the effect of different focal lengths when using peripheral vision or just moving our eyes up and down as we do all the time without thinking about it. As stated by cdayton, we are also used to see converging lines with out eyes. We are so used to it in a 3D world that manipulating the image or the lens to keep the lines parallel in the image causes the tops of buildings to seem unnaturally expanded. To mitigate the problem of using flat image planes to cover varying distances in the same photo and while trying to prevent convergence, I try to be as far away from a building as possible and use a longer lens. Not always easy where objects near the building are blocking the view.

You have the same perspective keystoning with your eyes. It is called "perspective". It started being recognized by artists in the 14oo's and the paintings became much more realistic as a result. Things look smaller, the further away they are from you. That is nature. If they don't look smaller, the image does not look natural. If you and your camera are on the ground, the bottom of the building is closer to you than the top and in nature looks smaller. So why try to get rid of a perfectly natural and real fact of geometry?

Convergence of vertical and horizontal lines is a natural visual phenomenon. Your camera sees what your eyes see. Some correction for certain architectural subjects is acceptable and makes the image appear more as our mind’s eye sees it. But trying to correct the convergence on an image of a skyscraper, for example, would result in an image that is unacceptable, IMO. It is odd in a way that horizontal convergence is acceptable but we seem to feel that vertical convergence needs to be corrected as a defect or shortcoming in an image.

Stan