IIRC one of the big differences/problems is that every one of them have completely different control rooms and control systems. To the point that you cannot train on one reactor and walk into another and operate it. If we at least standardized that side of things, the cost of operations would surely come down
I also question (and happy to be enlightened about) how much the local project needs some massive change. That a pipe might need to be a few hundred yards longer to reach the river surely isn't so consequential that you can't reuse designs
These things are built on huge sites that could be completely leveled and turned into identical square clean pieces of dirt with massive empty land around them How much variation can there possibly be that can't be handled outside of the critical reactor design area
Then again, perhaps it is one of those things where you actually need to make a few of 'em to gain experience.
Guess that's where they took it from.
Skimming the paper it seems a large part of the issue is that it's difficult to mass produce something which needs to be integrated into a variable environment, and attempting to do so is detrimental compared to custom solutions.
This seems to mirror what I've found in the software world, where a custom integration is often much cheaper and easier to maintain than trying to manhandle some generic library/software to do what's needed.