I think the main reason is that it is an easy way to get the edges within tolerance of the input curves, because
the input curves might be messy.edges
BlendSrf does the same and some surface creation tools inherit the knot structure of the input curves
Extrude tapered often makes fully multiple knots.
Offsetsrf and shell used on polysurfaces will make both multiknots and fully multiple knots, and are probably the most likely to produce unusable garbage due to those commands heavy reliance on inserting multiple knots. Offsetsrf used on surfaces can be prevented from creating multiknots by using the loose option. If you need to increase the accuracy of a loose offsetSrf you can temporarily add knots where needed to the input surface
The main issue is added unneeded complexity which can lead to downstream problems. Its a whole lot more efficient to simply avoid problems than to wait until you discover them later when they trip you up. The internal discontinuity of multi-knot surfaces is unnecessary complexity in itself but it also results in further unnecessary complexity and further degradation of continuity when the surface is used in additional modeling. For instance when you try to make a sweep 2 from trimmed edges of an existing sweep2 - that is when things can get really ugly when you don’t remove the multiple knots.
I have been doing it for more than a dozen years and have never seen a case where there is a downside.
I don’t doubt that someone can make input surface edges messy enough that the multiple knots will become necessary to make the edges match, but that is the downside of making a mess not the downside of removing multiple knots.