I ran into the same frustrations with large data sets. Work around which works rather well was to breakdown the data sets being transferred for sychronization. In our case we send packets of 1000 records and speed issues were resolved.