Performance tuning is a wide topic. The Performance Tuning guide for PowerCenter gives some fairly good guidelines how to start analysing performance trouble. The first and second step should always be to check for so-called "source bottlenecks" or "target bottlenecks"; if these are ruled out, you can continue to analyse whether you have some "transformation bottleneck".
Please browse the Performance Tuning guide for these first steps and follow them. It doesn't make much sense to try improving the performance of the UDT if, for example, you have a target bottleneck.
If you have enabled the event log property while deploying, try to disable it. Performance can only be improved if we identify where (transformation) there is an delay. May I know the Input/Output statistics? 2000 rows/min per partition doesn't seem bad.