The terms are used in different ways in different contexts I'm going to define them:
- centralised computing: every device does its own computing
- distributed computing: device sends off as much computation as possible to other computers
For example, to supercomputer users "centralised computing" means sending jobs off to the supercomputer, but you're talking about doing computation on your computer rather that off in the cloud.
In general, distributed computing is far, far more sustainable than centralised. This is mostly because computing needs these days go right down into tiny trivial devices that you might not even be aware of. For example, all the security cameras that have automatic number plate recognition built in, or smart thermostats (that react to outside temperature, people being in rooms or power cuts, for example). Taking a very simple device that costs a couple of cents and uses a few microwatts and bolting a network interface on it will make the interface into the dominant cost and power sink.
Even for larger computations, it often makes sense to use the computing power that's available rather than sending tasks away. So your smartphone camera will do face detection locally, but when it needs lots of memory and processing power for speech recognition will send that to a remote computer.
That illustrates the crossover question you seem to be concerned with: at what point does it make more sense to send the task away? And the answer varies depending on the complexity of the task, how easily it can be sent away, and what other computing resources you have available.
One core problem is the cost of securely communicating with the distributed computer. That takes computation itself (for the encryption) and often means redundant communication channels (when reliability matters) and communication takes energy too (a battery powered device using wifi is probably the example people are mostly familiar with). Any of those can be enough to make local computing the only viable solution.
The converse is also important - what does it cost you to safely make your computer available to others? Things like data security, power cost, wear and tear, and your time all factor in. You're not going to be happy if someone starts mining bitcoin on your smartphone, for example, even if you consented (on page 500 of the T&C for a free app). But if you offer your home desktop to a community project but never leave it on, or connected to the internet, you are just wasting whatever effort it costs them to add you as a contributor.
But that's not what blockchains like BitCoin do. They address an entirely different problem: how do people verify the actions of others? There are lots of uses. The computational cost is a side effect, what blockchains do is allow you to be fairly sure that what someone tells you is actually true. Money is the obvious answer - you don't care too much whether it costs a few cents to find out whether you'd actually been given $1000, but you care a lot about whether you have actually been given the money.
Where that effects efficiency and sustainability is when the cost of building the blockchain starts to dominate the process, or is much more than the cost of alternate methods. A lot of tasks don't need the high level of security that blockchains provide, for example, or the blockchain just adds a layer to an existing process that only needs it sometimes, or they need that level of security, but at times that blockchains will not work. That's why we don't blockchain instant messages or HTTPS request but we still have paper money (it even works underwater!)
Imagine an email forward chain, for example. It starts with something important - your government sending you a court order. You forward it to your lawyer (also important) but also to your friends (no-one cares) and the media (possibly important). Only the important cases really need a new entry in the blockchain, everyone else just needs to be able to verify the blockchain. So computing blockchain entries there is a complete waste, however "efficient" the computation is.
