Full table scans do not necessarily reduce undo generation. This is only true if the procedure commits after every n rows processed, and if more than one row will be affected in many of the blocks. If so, almost every second or subsequent change to a block will require extra redo and undo to clean out the block and allocate a new ITL entry.

While going through the Questions and Answers section on your web site, under Rollback Segments, you mentioned that full table scans generate less undo. Can you elaborate on this?