If Intel produces a quadcore with a defect core. it goes to the trashbin.
There is some redundant functionality that is disabled which is used to account for random errors (particularly in cache) but even if the entire core is broken then you still perform die harvesting where appropriate.
AMD sells triple and dual core GPUs which are die harvested quad cores with 1 or 2 broken die. (although as their manufacturing improved and demand rose they started disabling perfectly good die to meet quantities)
GPUs contain basicly large amount of "copy/paste" logic. And you can easily disable one without sacrificing the product too much.
GPUs do contain a lot of copy/paste logic, but your assertion that this is why they are built larger (due to being less susceptible to manufacturing errors) is untrue.
AMD actually took steps towards a focus on making them smaller in order to avoid manufacturing defects to increase yields.
There is strong competition in that field which justifies pushing the envelope. GPU performance increase per generation are massively bigger than those of CPUs.
Beyond the competitive pressure though there are various reasons to it a big one is the multi vs single thread performance as well as the cost to design.
Designing a CPU that adds more transistors to do tasks more efficiently is difficult, expensive, and time consuming. It is a lot easier to add more "cores" for work in parallel... its not as efficient a design but it is a practical compromise in R&D.
It is all about R&D cost, a very advanced single core processor using same amount of transistors has been hypothesized, as well as methods of building it (intel is actually financing such research) and it will probably outperform the "add more cores" designs but it just takes too much R&D to make those; maybe when die shrinking hits a wall we will see those crop up. Instead we have many cores with each generation bringing slight improvements to each individual core which are then replicated across the entire chip.