Hmm, I’m not sure, but this could be a limitation of the “cluster” function. From what I understand, the cluster has to have the same datasources configures as its nodes - but doesn’'t use them. An optional separate IP address is probably used for ping checks and could also be used for other active checks.
Perhaps it helps to configure the vSphere Cluster as a normal monitoring host and enter the ESXi hosts as parents.
In my environment, I usually query the vCenter (not the ESXi hosts, not the vSphere cluster) and have it map the information to the hosts via piggyback.