In AbInitio ETL tool, If there is a inner join component and million records are matched on some keys, there is some simple transformation logic in the join itself and rejected records are collected to process further.
Will join take same time to process the records if they are matched compared to unmatched?
In general, as with everything, run a test, and see.
If you think about it, having records joined should run faster as compared to no joins. Reason being is that once you join, you can discard the (inner) records, and it will not be prescient in the next iteration of trying to find a match. Resulting in less computational workload.