From my point of view :
Averaging the (host based) ready values for e.g. 100 vms on an esxi hosts is pointless.
That would be like averaging CPU load in top over all processes…
It should be either the real values of each vcpu/vm or at least averaged per vm, if the first option is not available.
These values would support us in CPU constraint troubleshooting and you could set sane values for rulesets. A general rule of thumb is that cpu ready should not exceed 5% per vcpu.
Actually it should be possible to retrieve values for single vms instead of the (whole) esxi and calculate the summation into the percentage values.
A good explanation is here :