I don’t see how selective replication should have any special effect on view indexing.
But if I understand your posting correctly, you’re actually suffering from two problems: Replication takes very long to complete and view indexing puts a high load on the server. Both issues are often caused by the same reason: All or many documents are deleted and recreated (e.g. by an import agent) in one of those databases. This is never a good idea. This way all the old replication stubs plus the new documents have to replicate and normal automatic view indexing can become a performance bottleneck.
If the overall number of documents is high enough, selective replication will not have a big influence.
A more exotic reason could be some scheduled code that deletes the replication history. But that would not account for increased view indexing activities.
There are no big imports happening on the database. It will just be users creating/editing/deleting documents.
When a database uses a formula to replicate, what view is used to run the selection criteria on? Would the database not index its views to ensure all documents are replicated?