On my two databases, the buffer cache were way too small. One buffer was 128 mb, which was the default value set in db_cache_size; one is 64 mb while the db_cache_size value is 0.
On both servers, the memory_target parameters were set (5g and 3g respectively) while sga_targets are set to zero. It seemed ASSM mis-configured the buffer cache.
After I set the sga_target to a reasonable value, ASSM began to adjust the cache buffer and set them to a much higher value. The first one was set to 700 mb and the second 608 mb.
Both databases are 220.127.116.11 on solaris 10.
Maybe we should either set the sga_target, or set a high value in db_cache_size.