I have to really regret I became regular reader of your blog just two months ago !!!. What a great post, never read by me before. What a pity!!!

Apparently, it is also giving some more answers to our last QUIZ NIGHT :)

Sorry about the delay in replying to this – I think it got hidden in a backlog.

The SQL to create the table and indexes looks like this:

create table t1
as
select 
        trunc((rownum-1)/15)    n1,
        trunc((rownum-1)/15)    n2,
        rpad(rownum,180)        v1
from
        all_objects 
where
        rownum <= 3000
;


create table t2
as
select 
        mod(rownum,200)         n1,
        mod(rownum,200)         n2,
        rpad(rownum,180)        v1
from
	all_objects 
where
        rownum <= 3000
;

create index t1_i1 on t1(n1);

After you’ve created the data and collected the statistics, the SQL to query the data looks like this:

select 
        count(n2) 
from
        (
        select
                /*+ ordered use_nl(t1) index(t1) */
                t2.n1, t1.n2
        from
                t2,t1
        where
                t2.n2 = 45
        and     t2.n1 = t1.n1
        )
;

Run the query a couple of times before and after the index rebuild to deal with any cleanout.

I’ve been trying in vain to reproduce the issue (looking at “consistent gets” from autotrace and from 10046 level 8 traces in different runs) but do not get a noticeable difference after a TRUNCATE and REBUILD. {I’ve taken care of delayed block cleanout as well).
I get the same number or a slightly higher or a slightly lower number of “consistent gets” after a Table TRUNCATE and/or Index REBUILD on thee driven table in the NL join.