Major Description
We can distribute load across multiple active transaction manager instances, if we can partition the update key space such that each partition is mapped to a given transaction manager instance. As long as updates do not span partitions, this will be safe as the transaction manager will preserve atomicity and ordering within the partition. This is somewhat like the HDFS federation strategy.
The easiest place to start would be providing support for mapping tables to a given transaction manager instance. If a client attempts updates that span multiple partitions, the client or manager should error out.