answer to 1) To process the record sets , I scan through the cursor of A-table keys; for each key, I SQL the records with that key from the 3 cursors I extracted from the remote data source. From the A-table cursor, I get one record; from the B-table cursor, I get 1 or more records; from a third (C) cursor, I could get up to 20 or more records.
answer to 2) I don't see how it would help. If i limit the number of records retrieved from a cursor, how do I know I have them all, since I do not know in advance how many B or C records I will get. If I get them in groups, or even one at a time, I would have to collect B and C records until the key changed, put the record(s) with the new key aside, process the records with the old key, then read more ... yada yada yada. Been there, done that; it's what high level languages and object orientation are supposed to save us from.