I was also considering the same question. My eventual use of NSX will likely drive it to be a multi-PSC/F5 design so I can use NSX cross-vce
I assume you are planning 2000 desktops per block within your pod. If so I assume you'd have to setup a Horizon pool of 2000 desktops per block. One pool = one vCenter. What baffles me is if all desktops in each block are the same how does one automatically load balance a connecting user to a pool with the least amount of active desktops?