This is all just surface. There are not enough lanes on consumer boards. If there were more lanes/better slot layouts, you could have all these features. But when it comes to actually significant features, like running multiple GPUs for local LLMs, you are out of luck.