For now, I don’t have a clear picture on NLJ batching and prefetching. I would do deeper research sometime myself, but still expect that some gurus would unveil it in the near future.

Could you elaborate on the mechanics of the NLJ-batching?
How is the improvement reached?

regards Hans-Peter

I can reproduce it on my 11.1.0.6 and 11.2.0.1 database on Windows(32bit).

Could you change the statement like following? (broaden the range of the returning rows)

select * from (
	select /*+ leading(t1 t2) use_nl(t2) index_asc(t1) index_asc(t2) */
		rownum as rnum,
		t2.c1,
		t1.c4,
		t2.c2
	from t1, t2
	where 
		t1.c3 = t2.c1
		and t1.c1 = 1
		and t1.c2 = 0
		and rownum <= 1000
) where rnum >= 1
;

Which version are you using ?

I cannot see the behavior on 11.2.0.1 on Linux