The CPU is used by these queries, there are a lot of buffer gets and the top wait event is latch: cache buffer chain.
The problem is that in other environment with Oracle 11g (11.1.0.7) and 10g these problems doesn't happen.
My question is if the optimizer of 11g can change the execution plan, even if the statistics doesn't be updated. And in one moment to another the Oracle may be resolve the execution plan in the wrong way - is just a guess.
The most rare is that these are optimized SQL statements.

Any idea?

I can't send the AWR (privacy policy), but if you need some part I will try.

Thanks,
Federico