Two weird performance problems

I have a set of queries joining two copies of what should be the same 10GB-ish table to test for duplicates, anomalies and missing rows between the two. If the table order is t2 -> t1, onstat -u shows a burst of disk read activity but this soon comes to a virtual standstill. The table is far too big for this behaviour to be accountable by buffer reads. I've eliminated stuck logs and checkpoints too. If I force the order t1 -> t2, it works fine and the diskreads gallop away. The equivalent index is being used in each case. I can reproduce the problem as many times as I like. oncheck -cd and oncheck -cI haven't found any problems with either table. Another problem we are having is with painfully slow ontape archives; the time for a level 0 has deteriorated over the last few weeks from 8-10 hours to 40+ hours. The tape drive light only flickers briefly every couple of minutes. I know there are ontape performance problems with very old pages but I can't help wondering whether these two problems could be related, especially as t1 is quite a bit older than t2. I've tried restoring the instance to an identical, spare box and the 'slow index' problem happens just the same, but a level 0 archive completes in 1 hour 28 minutes. Any pointers anyone? Thanks Andy Kent