Imagine two people buy an acre of land, one of them builds a skyscraper on it, the other a garden shed, by the logic you're using you would view the two as exactly the same, simply because they each have an acre of land.
You don't know how large the pixel sites are, and as they grow larger they become closer to each other.
Imagine that the two sites have no fences, people from each site are free to pass into each others land, but it's further to go from the shed to the boundary. Imagine a simple wire fence, everyone can see into the neighbouring plot and some are able to climb the fence into it. Now imagine a huge concrete wall no one can see it, and no one can pass through. This is isolation, and it's very important, because you can have small plots where little can pass through and cause noise.
Haha, flake... that was one of the most incomprehensible similes I've ever seen :-) And it had no cars in it. But I agree that larger pixels don't necessarily translate to higher S/N. What mostly matters is the number of photons captured. Whether you capture them on one or four pixels doesn't matter. The exception would be in the very low S/N domain, where the noise becomes read-out dominated. Also, since the well capacity scales closely to the area of a pixel (not volume!), the dynamic range does not change much with pixel size either (the well capacity will be smaller for smaller pixels, but that is exactly made up by there being more of those pixels for an equal area).