Matt Kaberlein raised these questions today, which seemed like good ones to me:

Just wondering if the following are still good best practices guidelines at this point:

• Dedicate specific datastores to specific Oracle data types and/or Oracle DBs?
• Map VMDKs to separate datastores to increase isolation and performance parallelism?
• Leave roughly 20% of available capacity on each datastore for ESX overhead if applicable (vMotion, VM snaps)?