There's no reason to ditch it entirely, they've replaced most of its components over time so it's something of a ship of theseus at this point. People continuously insist that they're being held back by its technical limitations. I've specifically been told that their engine can't handle ladders in the past, yet somehow those made it into Starfield. Not to mention how they managed to add functioning netcode and client-server architecture to the engine for Fallout 76. Most people saw 76's release as an absolute shitshow (and it was) but anyone who has worked with netcode would appreciate that it's a miracle that it worked as well as it did, because nothing they started with was ever designed with network synchronization in mind at least for over a decade.

If they took something like Unreal and modified it so they can do the same things as with their current engine they'd likely run into similar performance issues if they were attempting the same scope. All of the stuff like persistent objects in cells that are essentially signature to their games (instead of just having a few interactable objects in each cell) will be somewhat computationally expensive no matter what engine they're implemented on.