What if you need to reliably lock down a particular operation to a client on a single computer? Behind a NAT? Perspective can't do it. Vision can. Vision can even probe its environment for things like motherboard serial numbers, if you have that need.
What if you need to reliably handle a barcode scanner or RFID reader, or similar device, no matter what page the client is looking at? Perspective can't do it. Vision can.
What if you need to use a physical security token with certificate or with a challenge-response API? Perspective can't do this. Vision can.
Do you have a hard requirement for some kind of machine jog with momentary pushbuttons? That needs to reliably release? Perspective can't do this. Vision can (not really out of the box, but possible).
I hope that some of this will be integrated into Perspective Workstation and Mobile before too long. But for now, Perspective has some gaping holes that are commonly used in machine HMIs.
One other missing piece: Vision can launch local applications of any kind, monitor their operation, exchange data with them, and kill them off. Perspective cannot do this. And likely never will.