MultiFetch can be a very significant factor in performance. Suppose you can reduce Adabas calls from 10,000 to 200; you would see a sharp drop in elapsed and CPU times.
If you have a file with 30,000 records, and need 10,000 of those records, and need them in sequence by SD-99, it might well be the case that reading the file in physical sequence and ACCEPT’ing records in the desired range, and then SORT’ing the resultant 10,000 records will be more efficient than any READ LOGICAL.
There are many criteria you would have to take into account; e.g. average number of records in a physical block, how close the file is to SD-99 sequence, whether UPDATEs will be performed (also critical for MultiFetch considerations), etc.
There is NO simple answer for questions such as you asked. There are too many variables that go into an answer.