Thanks to Frank Denneman and Duncan Epping and their book vSphere5 Clustering Technical Deepdive I have recently learned a lot regarding the changes to clustering in vSphere5 but also something I was not aware of in previous versions of ESX(i). Storage vMotion is affected by the block size of the source and destination datastores; Only if the block size is the same on both will vSphere 4 use Changed Block Tracking, when the block size of your VMFS datastores are different however, ESX falls back to the legacy snapshot based storage vMotion used in ESX(i) 3, which has explained why sometimes (for me at least) storage vMotion can take a lot longer than you think it should. A good point to note then is that the block size come vSphere 5 is always 1MB, as long as the datastores are created new and not upgraded, in which case the block size remains the same. A second boon is that in the new version of vSphere VMware have introduced yet another storage vMotion method (not surprising given the leverage of datastore clustering) using a mirror driver, which will quiesce the VM once to inject the mirror driver into the stack and simultaneously write to both datastores and quiesce the VM a second time to move datastores. One of many many improvements in the new version, vSphere 5 is looking very interesting indeed.