Back to partitions, I've read about sharding today, the post was written by Vitalik and it describes exactly what is being implemented. It said clearly that if a call from one transaction group is made to another transaction group it would produce "out of range" exception. Its not clear to me why you even discuss the possibility of such calls that would be executed.
The second thing I did I watched an interview with Vitalik where he said that in Ethereum 2.0 they are thinking about getting away from every node executing every transaction. In a paper you quoted before exactly such solution is described with double decker blockchain where one layer is used to simply store the state. This seems to be the exact solution they are going to adopt.
need_war you need to understand that there are no problems in computer science that can't be solved. If particular architecture doesn't work they will change it to the one that works. The whole thing about blockchain with all the excitement around it seems to me like we say in Russia "sucked out of finger" . But that doesn't mean that Ethereum is not gonna work and that we won't be able to make money on it. Crazy youngsters like Vitalik will develop lots of shit on Ethereum just because people have nothing better to do. No point to talk further, what I studied so far is enough for me to start pouring money in mining farm for it.
Than they need to come over the CAP theorem - could be a bit tricky.
'Proof' is there:
https://bitcointalk.org/index.php?topic=1319681.msg13490710#msg13490710 
Yes applications on Ethereum will not work good for everybody all the time, but so what, it will work most of the time for the majority of participants.
Yep, so all APP Builders with big assets (& risk) behind, will never use ETH as is, rather fork it to private use or use some MySQL or NoSQL stuff...
Edit: BTC is still seen as a risky experiment, so pls consider ETH as even a more risky one. Just do own risk management for market AND cyber risks.
Here, a fork of Factom seems to be used:
http://www.coindesk.com/blockchain-startup-aims-to-secure-1-million-estonian-health-records/I've spent most of my life in software since 1996, right after the university I was in the team that created first DSL modem in the world, later I worked for a company that was working on powerline modem. Many big companies with lots of money spent fortunes trying to get it going, at the time everybody thought that its simply not possible. After all a company called Intellon actually made it work with pretty simple idea that was the winning one. Today most of the problems have been solved in software and Ethereum will get it right sooner or later, if they won't may be TPDB_need_war will tell them how to do it

Bottom line it will be working, just a matter of time. All they have to do is to slightly modify the original idea that everything has to run on a blockchain. There is absolutely no reason to stick to blockchain when there is no need to. Blockchain is not the only answer to "decentralized freedom" it will have to evolve into some hybrid monster that will do the job.