Correlated subqueries takes more time in $lookup

We need to do a lookup operation with the collections having more than 1 million records when we use the pipeline inside the lookup takes more time than splitting the pipeline into sub-stages. Is there any reason behind it or in what situations we can use the correlated subqueries operation?

– pipeline
– $match
– $unwind
– $project

Note: Necessary indexes have been created for the lookup fields.