I just ran into a similar issue. While only calling the select statement 2 times, each time had so many entries in the 'for all entries' list, that the compiler converted this into about 700 calls to the select. Now since the select joined three real tables on the database, the trace shows this one select as being the slowest item in this application.
I think that happened because 'for all entries' gets converted to an 'IN' clause, and then the total number of characters in any SQL statement has an upper limit. So the compiler must make the select statement over and over until it covers all entries in the 'for all entries' list. Is that correct?
Since every database I ever saw has the concept of db temporary tables, I have used db temp tables many times for this sort of thing.
The ABAP compiler could determine that more than one IN statement will be need, then use an alternate: write all the FOR ALL ENTRIES to a db temp table, then join on the db temp table, then drop db temp table. Since the compiler does this sort of thing, no application code needs change to get the speed boost.