What causes cluster wait time? How do you resolve?

 We have a simple update statement that has performed terribly when run in a RAC database.  An ADDM report shows 3 different findings related to this call.

1) SQL Statements consuming significant database time were found (the update ran 27177 times with average elapsed time of 0.52 seconds per execution)

2) Read and Write contention on db blocks was consuming significant db time (inter-instance messaging was consuming significant db time, Wait class "CLUSTER" was consuming significant db time)

3) Sql statements responsible for inter-instance messaging were found (update ran 27122 times, average time spent in cluster wait events was 0.44 seconds)

What causes the cluster wait time to sometimes be so bad?  How can we resolve the issue?