The only reasons I can think of involve handling cross traffic, both at intersections and when pulling out of parking places. And even that could probably be largely solved by replacing the main front camera with a higher-resolution front camera so that you won't need the narrow center camera, then replacing the narrow and wide front cameras with steeply angled cameras pointing off towards the sides. That approach would be an easy retrofit, too.
How do you know it would be an easy retrofit?
If they design it to be an easy retrofit, it will be. That's not to say that they can't make design decisions that would make it a nightmare, of course.
Realistically, any computer that they replace FSD with isn't going to be able to draw a lot more power than the current one, because doing so would adversely affect your driving range. And we don't have to worry about an upgraded computer having dramatically higher thermal output, because that would require drawing considerably more power.
Chances are, new wiring will be required, but we're just talking about running wires from the location of the FSD computer up through the dash and around the front windshield, so that should be relatively easy unless there isn't enough space for extra wiring, and even then, they would have to build updated trim parts with a larger gap for new cars anyway, so it shouldn't be a huge cost to upgrade existing cars.
Also, 4K video currently requires a whopping five pairs per camera plus power, whereas the current cameras require only one, which could make it much less practical to conceal the wiring harness around the edge of the window even for new installations. (Imagine four full Ethernet cables here, with two around each side of the window, and that assumes that they don't use 4k for the interior camera.) So I wouldn't be surprised if they end up multiplexing the signals over optical fiber (likely one fiber per camera), which should easily fit into the space used by the existing wiring harness. And then take optical inputs directly into the HW4 board to avoid any extra conversion steps.
And realistically, they will probably upgrade all of the cameras. The existing cameras are likely to be EOL soon, because they're fossils. So that would necessitate that the hardware support both the old-style cameras (that can't feasibly be upgraded) and the new ones. This might require adding some extra conversion hardware to convert the older cameras's outputs to feed into whatever inputs HW4 uses, but nothing too complex. Or make HW4 have multiple input boards, one of which is all-optical or all-four-lane-MIPI or whatever they end up using for the 4K cameras, and one of which is upgraded for only the main cameras and uses the existing single-pair LVDS connections for the other cameras. Many approaches exist for making upgrades straightforward.
Of course, if we were talking about upgrading every camera (not just the front/center trio) or adding cameras in other places, that would be another matter, but that would also be well outside the scope of what I said would be an easy retrofit.