As pointed out above, this is mostly relevant for inner loops.
That said, there is a healthy culture of micro-optimization one-upmanship on this forum, which occasionally runs amok, so it is easy to get the impression that there is a standard “bag of tricks” that is absolutely necessary write fast Julia code.
I prefer to think of these this the opposite way: idomatic Julia code is reasonably fast out of the box, provided some pitfalls are avoided. 90% of the time it is not worth optimizing it further. The 10% should be identified by profiling and benchmarking.