I would be checking for blocking. Replication is attempting to deliver changes to about 10% of the records - to me that sounds like you probably end up with replication having exclusive locks on most of the table(s) in question.

Since you have a slow connection, have you looked using a different profile for the distribution agent. One of the changes that you can make is to decrease the bcpbatchsize. From memory it is every 100,000 records. On a slow network could take quite a while. I would decrease this to something like 10,000.