I just thought I would through in my hat into the ring here.

When new releases come out I always check and see how fast they NSum and NIntegrate for the sake of computing digits of the MRB and MKB constants, respectively. Sometimes the speed slows down a little, due to what I think is the cost of Mathematica having more overhead to deal with (bloatware if you please), Nonetheless, over a long period of many releases, the speed does increase.

I can see that Mathematica does take a balanced approach to adding new functions: they consider what speed the new functions cost.

Neither approach is better than the other.

People are cognitively different from each other. Each person's mind works differently from each other person's. Some people's minds will work better with many large named pieces of functionality. Some people's minds work better with a smaller number of more granular functions as part of a tighter core language.

To be specific: People with a high inductive reasoning aptitude, an ability to match functions with needs, who are very practical and/or who find learning new functions easy will probably do better with a language having many named larger pieces of functionality. People who are more creative, have a high deductive reasoning aptitude and/or strong linguistic, cause-effect reasoning or visual skills might find a tighter core language easier to use.

I think the poster and myself belong more the second group and Stephan Wolfram belongs more to the first group. That is fine. The world needs all types. A language that supports both types of thinking would probably be most ideal as long as the two natures of the language are well integrated. It comes down to cognitive differences among humans as to which approach is better for which person.