Parallel replication works best on workloads that meet the following criteria:
Data are stored in independent schemas. If you have 100 customers per server with a separate schema for each customer, your application is a good candidate.
Transactions do not span schemas. Tungsten serializes such transactions, which is to say it stops parallel apply and runs them by themselves. If more than 2-3% of transactions are serialized in this way, most of the benefits of parallelization are lost.
Workload is well-balanced across schemas.
The slave host(s) are capable and have free memory in the OS page cache.
The host on which the slave runs has a sufficient number of cores to operate a large number of Java threads.
Not all workloads meet these requirements. If your transactions are within a single schema only, you may need to consider different approaches, such as slave prefetch. Contact Continuent for other suggestions.
Parallel replication does not work well on underpowered hosts, such as Amazon m1.small instances. In fact, any host that is already I/O bound under single-threaded replication will typical will not show much improvement with parallel apply.