Adam Back (Blockstream CEO and original cypherpunk) was the originator of the extension blocks idea, not Jeffreys and Poon. And Adam Back abandoned further pursuit of
his own idea for extension blocks, because it has problems.
Extension blocks creates a 2 tier network, and in a way Segwit does the same. But the Segwit 2 tier structure is far preferable to the 2 tiers with Extension Blocks.
Segwit allows for 2 block types: transaction blocks and witness blocks. The transaction blocks are just the same 1MB blocks we have now, hence why Segwit works as a backwards compatible soft fork. The new witness blocks that Segwit permits cannot contain transaction data, only the signatures for the corresponding transactions in the 1MB block.
So, the 2 tiers with Segwit are like this:
- Nodes that handle both block types, transaction blocks and witness blocks
- Nodes that handle only transaction blocks
So, the way to make your node part of the more complete blockchain (i.e. transaction blocks and witness blocks) is just to use version 13.1 or newer. And because using the new version comes with heaps of other benefits anyway, it's basically win-win, apart arguably from the extra large blockchain we'll have (but that's a compromise everyone knows must happen anyway, and it comes with it's own benefits, i.e. more on-chain transaction capacity)
With Extension blocks, the 2 tiers are like this:
- Nodes that handle both block types, the original 1MB blocks and all or some extension blocks
- Nodes that handle only the original 1MB blocks
The problem with this is that any number of extra parallel blockchains can be branched off from the main-chain using the Jeffreys/Poon proposal. When you receive a transaction from someone sending it from an extension blockchain, you must download the whole of that extra blockchain, upto the block the sender is sending the BTC from. Otherwise, how can your Bitcoin software verify the BTC is the real thing?
Because of this flaw, it can be argued that this idea is at least as bad as Emergent "Consensus" (what a misnomer that expression was), or perhaps even worse. It's maybe even worse for Bitcoin network decentralisation than the original draft of BIP 101 (20MB blocks with exponential growth) proposal.
If the extension block chains are unknown and unknowable in size and number, until you need to download a whole extra blockchain because you need to receive money from that extra blockchain, how can anyone ever know how big the whole set of blockchains actually is?
That's the real issue here. Extension blocks give miners the ability to create an infinite amount of additional Bitcoin blockchains, all potentially needed by the regular user. It gives miners too much power over the network, in a different way, but similarly to EC/BU would do. No wonder miners like it.