Well-written asynchronous code scales much better than synchronous code, because the former lets you use one thread to talk to e.g. 100 server simultaneously, while the latter requires one thread per server. So much is true for any language. However, for languages that do not support something like async / await, writing asynchronous code tends to become very difficult very quickly, which is why it is often avoided. Async / await lets you write code that looks synchronous but is then automatically turned into asynchronous code by the compiler.
A neat feature for sure, but bitcoinj already uses async IO and has been used in high-scalability scenarios like p2p network crawlers and BlueMatt's relay network.
Right, I haven't decided yet. BTW, what would you estimate is the effort write a minimal SPV client library from scratch?
Define minimal. bitcoinj is the result of years of work but that wasn't full time effort, and some of that time was spent on things like documentation, example apps, lots of unit tests, performance optimisations, implementing Bloom filtering on the C++ side etc. You could ask Maxime who did the C client as he did it more recently, though his code is less complete than bitcoinj.
