Sidechains also solve higher fees. You can have more centralized and highly liquid sidechains with low fees, and highly decentralized chains with high fees. This can still all be done with the same basic token.
I think you're making the same type-theory error that just about every Bitcoin maximalist does: conflating
blockchains and
currencies. The questions of:
1. Should all crypto be done on one blockchain?
2. Should all crypto be done on one currency?
are completely separable. Sidechains are a way of using one currency but using multiple blockchains; hence if we go the BTS/ETH/whatever sidechain route then we are still going the multi-blockchain route, just not a multi-token route. Metacoins, and some versions of Ethereum contracts, are a way of using many currencies but still being on one blockchain.
Sidechains by themselves are not magically secure. If a sidechain as currently described is only used by a few people, then it will be easy to double-spend. The challenge of scalability is not how to split up the applications - that's easy - it's how to split things up while maintaining a high degree of security for each piece. Sidechains do nothing in advancing this frontier, because they are a currency innovation and not a blockchain innovation.
Why would someone run a non-financial service on a decentralized computer? Seems highly inefficient for everything that doesn't require absolute security and trust.
Name registries for one. Tracking ownership of smart property devices as well. If they become really cheap, we could even start talking about decentralized forums and the like. People already write such software in python, which is a >1000x efficiency hit over assembly, so it's not at all unreasonable that people will value decentralization enough to take a ~200x hit in efficiency for them, particularly since the blockchain only needs to store a small business-logic core.