It's becoming less of an issue if you are on 2.4+ with more applications leveraging pxgrid v2. Ideally you have two pxgrid nodes, but you could also have 4 if you're on 2.4+ and pxgrid v2 is active/active on any pxgrid node. This means that failover time of pxgrid v2 is much improved, but still relies on the end application.
I do have significant caution with pxgrid v1 applications, and pre 2.4 deployments. Planning has to be taken with the application owners on the other end of the pxgrid integration if they are relying on ISE for this. Pxgrid v1 is active/passive, and if you decide to go upgrade a node where pxgrid v1 connections are being leveraged, there will be an outage while services have to detect the condition and start up on the secondary. During that time there will be a pxgrid v1 outage.
After the first pxgrid node has been upgraded, you will have active/active pxgrid v1, but one active node will be in deployment 1(pre upgrade), and the other will be active in deployment 2 (post upgrade). This can cause issues with subscriber applications since it's not a typical healthy condition.
I have no one solution for upgrading pxgrid, each environment has unique requirements and ability to bypass.
