We consider the use of a cluster system with a shared nothing architecture for update-intensive autonomous databases. To optimize load balancing, we use optimistic database replication with freshness control. We propose a solution to transaction routing that preserves database and application autonomy and a cost model to estimate replica freshness. Then we propose an algorithm for transaction routing that takes into account freshness requirements of transactions and replica freshness. We implemented our solution on a Linux cluster running Oracle 8i and performed extensive performance experiments using the TPC-C OLTP benchmark. Our results show that our solution outperforms existing solutions for typical transaction workloads.