Basically because you need 2 computers (client and server) whenever you activate any feature.
Yes indeed, plus all the internet routers in between, plus georeplication between data centers, plus monitoring to centralize all the information to give a global picture to engineers and management. And you also need energy and resources to host all the developers with all the perks of a top position in a tech company to monitor and maintain the infrastructure.
In comparison, SSB clients can use the Internet routers only for the delta updates which happen only once per piece of data and many of our developers have much more frugal lifestyles. If we use propagation algorithms based on physical proximity between smartphones, we can do away with the routers entirely. I have not tried to do back-of-the-envelope calculations yet but I suspect we use several orders of magnitude less energy and there is still significant potential for improvements.
A whole system comparative analysis of energy flow could be an interesting paper for the future!