Okay, thanks.

I was testing this all day now, but I don't see any improvements. I used TOAD to visualize the wait events and the whole graph was dominated by SQL*Net-waits.

I wish performance tuning was a bit easier... (still struggling with performance tuning for my exam)