Is this behaviour repeatable ?
Is the table defined to be in the KEEP cache ? Is instance 2 configured without a KEEP cache ?
Hemant K Chitale
Table is not defined to be in KEEP cache.
Looks like something else is using the cache on 1st instance.
Get AWR report for all 3 instances for the period of the issue, and compare them to see any activity on 1st instance you dont see on 2nd and 3rd.
You can also check global cache statistics. Run this before and after your query :
select name, value from v$mystat s, v$statname n where s.statistic#=n.statistic# and name like '%blocks received';
Please check any wait events causing for that instance
Did you check how many sessions in each instance, when this issues is occuring, and please confirm all database memory parameters are same accross the cluster.