My two cents -
1- First try with async sink, and if this can help you to reduce lag in P1 kafka topic.
2- Second- Use custom deserialiser and introduce isFilter flag and kafka topic name in the pojo, and in your operator when you detect a back pressure check the current record topic name and if it is secondary one set the filter to true and this way you can filter out the records from downstream pipe line.
The only issue is, it is not pause but complete drop.