Personally I don't know the reasoning behind those books and I don't claim to have any unique insight on this issue.
However, external merge sort can be implemented in exactly two stages, and neither stage contain recursive subdivision. In other words, in both stages, there is only one level of subdivision.
Choose number of subdivision N.
Stage 1: Arbitrarily partition input data into N subdivisions. Sort within each subdivision.
Stage 2: Create N FIFO queues, each feeding from the corresponding subdivision in the sorted order. Perform N-way mergesort (binary heap or selection sort) by peeking each queue's head and removing the smallest element, sending it to the output FIFO queue. Whenever an FIFO queue become empty it will feed from the remaining data from the corresponding subdivision. The sizes of FIFO queues are chosen according to available memory, but the algorithm still works correctly even if all FIFO queue has size 1.
As explained above, there is no recursive subdivision of the problem.
This is just an observation. I don't know whether book authors regard recursive subdivision as a key part of D&C.
There are many external merge sort schemes. The one described above is only one example. Those other schemes might make use of recursive subdivision. I don't know much about those so I won't be able to comment.