But that's why Swift generates slower code. Memory is cheap, also for Apple, although Apple would like to hide that fact.
1. They had experience on the type of app and you tend to do things better when you do it the second time.
2. Reading between the lines they were likely running an ancient version of Java from 10+ years ago.
They already admit in the article that JVM setup costs were a huge overhead during scaling operations. JIT might also incur a similar overhead at the same stage. So, not the language per se, but the runtime difference must have made a difference.
Compiled native code may not be too different between the two. But, in terms of memory management, Swift's ability to keep the heap in a smaller area might contribute to better cache utilization which might implicitly contribute to performance.
In the end, as you said, writing better code the second time might have made the greatest contribution of course. That doesn't necessarily mean that language change has made no difference.
Yeah the startup costs are an active problem. I do think though that if they weren’t running an old runtime they wouldn’t have had the same GC issues.