Micro-optimization is only important if the numbers say it is.

The requirements that you are developing against should have some specification of performance data, if performance is at all an issue to the client or user. As you are developing software, you should test the performance of your system against these requirements. If you aren't meeting performance requirements, then you need to profile your code base and optimize as needed to reach the performance requirements. Once you are within the minimum required performance, there's no need to squeeze any more performance out of the system, especially if you are going to compromise the readability and maintainability of the code any more (in my experience, highly optimized code is less readable and maintainable, but not always). If you can get additional performance gains without degrading the other quality attributes of the system, you can consider it.

There are instances, however, where performance is of the utmost importance. I'm mainly thinking in real-time systems and embedded systems. In real-time systems, changing the order of operations can have a huge effect on meeting the hard-deadlines that are required for proper operation, perhaps even influencing the results of computations. In embedded systems, you typically have limited memory, CPU power, and power (as in battery power), so you need to reduce the size of your binaries and reduce the number of computations to maximize system life.