VMware Support Insider has just posted an article about a workaround previously posted on Virtual Lifestyle:

We have become aware of a number of customers reporting various seemingly random behaviors on their ESXi hosts, specifically on hardware in which OEM CIM providers have been installed.

You may not be able to power on certain virtual machines. Other virtual machines may enter an invalid state randomly. You might also encounter failed vMotion migrations at 82%.

Your first reaction to this might be to restart the hostd daemon, but in this case the problem does not go away, at least not for long.

Knowledge Base article 1035564 – Virtual machines in an invalid state fail to power on with the error: FoundryVMDirectlyOpenSocketToVMX: Failed to create socket pair, has been written about this issue, and will help you determine whether this is the problem impacting you, and how to deal with it.

While VMware works to create a permanent fix for the problem, there is a work-around which involves stopping the hardware monitoring process sfcbd (details in the KB). Depending on the system workload, this change may only temporarily resolve the issue.

Thanks to Support Engineer Solomon Wen for creating this KB article.