Cluster Resource Allocation "Memory - Total Capacity" is "Total Cluster Memory" (what you see in Summary Tab) minus approx. 2576MB of RAM reserved for each ESX host.
So if I have two ESX hosts each with 8GB physical RAM I can see 16GB Total Cluster Memory in Summary Tab. However I have two ESX hosts which has together reserved 2 x 2576MB which is approximately 5GB of memory reservations. So in Cluster Resource Allocation I have 16GB-5GB which is around 11GB of RAM.
The same should apply to Cluster Resource Allocation "CPU - Total Capacity". Each ESX host has reservation of 2341 MHz.
So if I have two ESX hosts each with 10.636 GHz I can see 21GHz Total Cluster CPU Resources in Summary Tab. However I have two ESX hosts which has together reserved 2 x 2421MHz which is approximately 4.8GHz of CPU reservations. So in Cluster Resource Allocation I should have 21GHz-4.8GHz which is around 16.2GHz. But during my tests I see there 18.4GHz which looks like only one ESX host reservations are subtract. What is the magic and why? Can someone comment it bellow the article?
If you want to know what amount of MEMORY and CPU reservations are reserved for particular ESX hypervisor component you can select some ESX host in the cluster and go to Configuration->System Resource Allocation and switch from simple to advanced view. You have to go through all components and sum all CPU and MEMORY reservations.
Cluster Allocation "Reserved Capacity" is sum of reservations of virtual machines and resource pools. Sometimes people are confused and surprised that reserved capacity is very high. That's usually because HA cluster is enabled and fail-over capacity is also reserved and not available for use.
So if I have two node cluster with N+1 redundancy and at least one protected VM is running then half of cluster capacity is reserved by HA.