Running such experiments can be educative and even addictive. However, keep in mind that microbenchmarks shouldn't be the only type of performance testing in your project. As Donald Knuth famously said:
We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil
This means that you should microbenchmark only code that matters, especially code on your hot path. Larger benchmarks, along with tracing and profiling, can be used to see where and when to optimize instead of guessing and optimizing prematurely. First, understand how your software executes.
NOTE: There's one more point we want to make regarding the quote above. It doesn't mean you should allow premature pessimization. Poor choice of data structures or algorithms, or even small inefficiencies that spread all your code, can sometimes influence the overall performance of your system. For instance, performing unnecessary...
