In the current design the "Total Reservation Capacity" value is simply the sum of "Used Reservation" and "Available Reservation". This is correct in normal cases.
"Used Reservation" includes the reservation configured at the direct child VMs and the child resource pools as well as resource reserved for HA Admission Control policy.
"Available Reservation" computes differently in normal cases and over-committed cases. In normal cases, it is result of the total available resource for reservation subtracting "Used Reservation" above.
When a DRS cluster becomes over-committed due to cluster member hosts number change (i.e. some hosts are moved out of cluster) the existing resource pool reservation settings is no longer valid under the new capacity(In this case, the "Insufficient capacity ..." message is displayed on the cluster Summary page for vCenter administrator to intervene and correct). In this case we would still want to allow VMs to be powered on at the cluster root node to reduce the impact from such invalid configuration. Hence the calculation of "Available Reservation" is changed to override the reservation configured on the child resource pools, which makes it a result of the total available resource for reservation subtracting reservation configured at the direct child VMs and resource reserved for HA Admission Control policy.
"Used Reservation" on the other hand remains unchanged in over-committed case. The sum of "Used Reservation" and "Available Reservation" actually counts in the resource pool reservation settings twice resulting "Total Reservation Capacity" larger than "Cluster Total Capacity".