DSXchange

Welcome aboard! :D

Start by defining precisely what you mean by "performance", the metrics you use to measure it, what values you got, what values you expected (and why you expected them).

Inspect the generated COBOL. It's not too bad, and in my experience reasonably efficient.

Have you isolated the "performance" issue to DataStage, eliminating something (everything) in Teradata as the cause?

Lookups are intended to be made against database (DB2) tables. In this case, the table is opened once and a cursor established based on the query - in non-mainframe environments this is called "prepared SQL".
DataStage then passes the search key value and receives a row in return.

I believe the problem arises from your design in using a sequential file as the source for a "get row by key" lookup. (Though I am surprised that the file is opened and closed afresh for each row processed. If you can verify this, please report it via your support provider - it's a hopelessly inefficient implementation.)