The script typically pulls in 1000 posts at a time.
It keeps track of what has been processed, so subsequent runs can skip data already run. I on scripts I’ve touched I also include an import_after setting that further speeds up subsequent runs by loading only recent data (also useful for testing with just a small subset of the data).
I would need to look more closely to see about whether posts are included in permalinks. They typically are not, but it can be done.
You’ll want all of your uploads on S3, so your backup will include only the database dump. You can’t really do anything to optimize that. You can either let discourse keep a certain number or tell it not to (or just set the number of backups to a big number) and let your rules handle it.