There is an article here but does not get into specifics on the performance of each option. http://aws.amazon.com/articles/4173109646282306
Oracle data pump is one of the fastest ways to move data between Oracle databases. However, the approach shown in the above article uses database links which can be very slow. A more performant method would be to use data pump file mode. Unfortunately, AWS Oracle RDS does not off access to the OS. There is a workaround to this by using a little known feature of the Oracle database called Oracle Managed File System
Oracle RDS has access to this Oracle Managed File (OMF) system. Once the file exported from the on premise database and moved to the Oracle RDS OMF you can use the UTL_FILE capability in the database and import a file into the RDS instance, and then do impdb (data pump import) from there.
Things to keep in mind:
1. Since the file sits on the Oracle managed file system, it consumes the space allocated for the database. So you will need to make sure you allocate enough space for the instance
2. Once you do the impdb operation, you can delete the file but till then it will consume space on the instance
3. File sizes will need to be less than 400-500 GB since the largest instance currently is 1 TB, and you will need space for both the file as well as the database
4. As you do the import, you will consume IOPS for reading the dump file and then writing to the database.