This article clarifies when virtual disk provisioning type can change on ESX hosts. (Disk provisioning type is sometimes called disk allocation type.)

Note: Virtual disk provisioning types are "thin", "thick" or "lazy zeroed thick" (LZT), and "eager zeroed thick" (EZT). 
For more details, refer: Virtual Disk Provisioning Policies

• After creating, migrating or cloning a virtual disk, you notice the provisioning type of the disk does not match what you selected in the UI or specified in the API call.
• After a VM has been running for a period of time, you notice the provisioning type of a disk attached to that VM has changed from thin or thick to eager zeroed thick.
• After snapshotting, restoring, or backing up a VM, you notice the provisioning type of a disk attached to that VM has changed.
• After inflating, growing, shrinking, encrypting, or otherwise modifying a virtual disk, you notice its provisioning type has changed. 

In general, virtual disk provisioning type is a mutable property. It can change without warning and without manual intervention. There is no guarantee it will stay constant.

• Such unexpected changes are especially common when NFS datastores are involved.
• Disk provisioning type should be seen as a creation hint. VMware software will make a best-effort attempt to create and preserve a disk as the requested type but this sometimes isn't possible.
• The only exception is for shared (multiwriter) disks on a VMFS datastore that are of eager zeroed thick (EZT) provisioning type. A shared disk on VMFS requires EZT provisioning, so VMware guarantees it will stay EZT as long as it stays on a VMFS datastore.

• It is not a bug and an issue should not be filed unless a specific VM operation fails with an error code because a virtual disk provisioning type has changed.
• The only known situation where a change in provisioning type can cause a failure is discussed above:
  • A shared disk on a VMFS datastore must remain Eager Zeroed Thick or it will fail to open.