algorithms: combination of fusion ordering with insertion ordering: temporal complexity

I am learning CLRS book algorithms on my own, without any help. You have an exercise that combines the type of fusion {O (n log n)} with the type of insertion {O ($ n2)}. He says that when the sub-matrices in the fusion classification reach a certain size "k", then it is better to use the insertion classification for those sub-matrices instead of the fusion classification. The reason given is that the constant factors in the order of insertion make it fast for small n. Can someone please explain this?

It asks us to demonstrate that (n / k) the sublists, each of length k, can be classified by order by insertion in O (nk) in the worst case. I discovered that the solution for this is O ($ nk2 / n $) = O (nk). How do we get this part OR ($ nk2 / n $)?

Thank you !