How about keeping a main blockchain, housed in several thousand, major repositories around the world, with mirrors and backups. Everything would go into the main blockchain.

There would be a smaller, secondary, "practical," everyday blockchain that would house the first record of every address, and the last two. Yet it would eliminate all the times an address was used between the first and the second last.

This way we would always have the full record when needed, but we would have easy access for everyday use.

Yes, it makes sense to remove/raise the hard limit from the protocol and let individual miners set their own limits since they are the ones most in touch with what the optimal parameters would be for their individual setups. If we went to a 20MB cap tomorrow I'd guess that no pool would even try to build blocks anywhere near that size given the high risk of getting an orphaned block with the current state of the network.

In the end, the whole block of transactions must be present on the blockchain at the most distant end of the network in 3 minutes to allow newly discovered blocks to be added upon it. Ideally, you need to transmit 20MB data in 1-2 minutes. Maybe it is possible to use multi-threaded P2P downloading to accelerate the data transfer

Blocks are just transaction data which has almost all already been relayed through the network. All that one has to send is just a tiny set of indexes to indicate which of the txn in circulation were included and in what order  and there are already alternative transports that do this (or even less-- just a difference between a deterministic idealized list and the real thing).  The data still has to be sent in the network, so it doesn't fundamentally improve scaling to be more efficient here (just a constant factor), but it gets block size pretty much entirely out of the critical path for miners.

Gavins proposal automatically scales 40% per year after the 20mb raise, then stops after 20 years - last time I read up on it.

After 20 years this puts us at ~16.000 mb or 32.000 TX/second. That's more than the VISA network and I think it will last a long time.

I also support Gavin and OP.

Great effort in posting. You hit the hammer right on the nail. But i think the title is slightly misleading for me. Perhaps a question mark at the end instead would suffice. I think a decision has to be made immediately because we are really close to reaching the limit with the current 1MB restriction.

Great post Death&Taxes.

However I have never heard anywhere people supporting to keep permanently the max block size to 1MB. So your post is somewhat misleading and is avoiding the most interesting questions. Which are:
1) max block size design : is switching from hard coded max block size 1MB to hard coded 20MB the smartest way to proceed ?
2) timing: is now the best timing to do that? It  appears there is no real argument to rush. Tons of innovation is going on, we are months away from a first side chain implementation. Perhaps in 6 months or a year from now we will have ideas we hadn't thought of.

From the comments here and on reddit, I don't see a real consensus on that subject. So imo it's better to keep discussing about it. We can reasonably expect to see concurrent proposals to Gavin's, with a different MBS design and perhaps a more appropriate timing.

For what it's worth: 

I'm the guy who went over the blockchain stuff in Satoshi's first cut of the bitcoin code.  Satoshi didn't have a 1MB limit in it. The limit was originally Hal Finney's idea.  Both Satoshi and I objected that it wouldn't scale at 1MB.  Hal was concerned about a potential DoS attack though, and after discussion, Satoshi agreed.  The 1MB limit was there by the time Bitcoin launched.  But all 3 of us agreed that 1MB had to be temporary because it would never scale.

Several attempted "abuses" of the blockchain under the 1MB limit have proved Hal right about needing the limit at least for launching purposes.  A lot of people wanted to piggyback extraneous information onto the blockchain, and before miners (and the community generally) realized that blockchain space was a valuable resource they would have allowed it.  The blockchain would probably be several times as big a download now if that limit hadn't been in place, because it would have a lot of random 1-satoshi transactions that exist only to encode information for altcoins etc.

At this point I don't think random schmoes who would allow just any transaction are getting a  lot of blocks. The people who have made a major investment in hashing power are doing the math to figure out which tx are worthwhile to include because block propagation time (and therefore the risk of orphan blocks) is proportional to block size. So at this point I think blockchain bloat as such is no longer likely to a problem, and the 1MB limit is no longer necessary.  It has been more-or-less replaced by a profitability limit that motivates people to not waste blockchain bandwidth, and miners are now reliably dropping transactions that don't pay fees. 

Are you saying that the original plan was to have no limit at all? What would happen if we were just to remove the limit now, instead of making it bigger?