OK Mr engineer.
So tell us why they need smart shift if there is no power limit and its the simplest , lowest bar to clear design.
Tell us why they are shifting load instead of making things simple and locking everything in.
I don't know for certain. That's the only valid answer in this whole thread, not speculation bandied around as fact.
Now, if I had to guess, I'd say it was to afford devs flexibility with game design. CPU-heavy game or scene? You can shift extra power to that unit. Graphics-heavy scene? Shift focus over there. You have a fixed transistor budget, and fixed die size based on yield expectations. You can only fit so many logic units in the design of either CPU or GPU, but you have the ability to modulate clocks to provide a boost.
Now that you have your processor design, how far can you push the operating range on your design, while remaining within an acceptable cooling profile for the packaging you want? That's where you set a limit on the frequency range, and thus current draw, which then determines the size of the PU you need to power it. This is how the first gigahertz processor was being designed when I interned on the IBM Spinnaker project. It's 22 years now, so I'm sure it's okay for me to talk about it. Besides, I think the project got folded into S/390 anyway.
Again, this is all speculation, and based on nothing more than conjecture. But I'm not pretending to be an expert on what the design decisions were with the console. I can only make assumptions based on my own experience working with a microprocessor design team. Power supply was never discussed in team meetings, and while I only worked on the top via routing design, I don't believe basing a chip design around something as simple as a PU makes any sense. And do understand that the "revolutionary" thing about Spinnaker was that it was supposed to be designed from the outside-in, meaning that you would start with the external interfaces, and make sure that everything down to the smallest logic units could comply with the target 1GHz clock. You're not building an electric car, or something that has the PU as the source of its functionality. The source in a console is the microprocessor, and you can spec your PU to meet the system afterwards.
tl;dr I'd say yields and performance would dictate the microprocessor designs, followed by cooling and packaging. Those would then dictate the PU. Although the PU would play some role in packaging, you can always make it external to hit your targets on the latter 2 points. Either way, PU shouldn't be the cost leader in the design. That should be the chips, memory, and new-tech SSD, so why let a PU hamstring you? These are professional engineers, so they wouldn't.