Every node in the network must receive and process every transaction in every block. There is no distribution or sharing of that load.
This isn't the part of Bitcoin that they are talking about.
They aren't talking about processing the data that they are receiving, they are talking about spreading to other nodes these data.
Nodes doesn't need to process data again and again for every other nodes that they connect to, and mostly not with every possible node.
Correct. And the fact that each node processes blocks makes the processing redundant but doesnt increase demands on individual nodes. I think Greg knows full well what I meant. It raises an eyebrow at the intentions of his posts.
Increase demands relative to what?

What "part of bitcoin" are we talking about? Think about what you guys are saying. If every node
did not need to receive and process every transaction in every block, and that load
was distributed, why would nodes be propagating data to other nodes at all? If a node is not self-validating, it is necessarily depending on trusted third parties, so any data it propagates is worthless (e.g. SPV). Self-validation is inextricably linked to the idea that peers are propagating data to one another.
The entire point is that self-validating and propagating each block to many peers is quite redundant, and requires far more resources than, say, downloading a web page.
Hence, epic meta commentary.
So now we went from 'massively' redundant to 'quite' redundant. Ok.
Yes nodes have to validate, that is a good point.
But we still get the multiplier effect where information gets propagated
exponentially. Also we have the relay networks. Greg is so fond of
pointing these out in the context of the fee market discussion where
it serves his arguments about the lack of orphaning risk, so why ignore
them here when they can help propagation?
Regardless, 2mb is still small when it comes to internet bandwidth.
(You initally made it sound like 2mb would have to be uploaded thousands of
times by a single node.)