A lot of different factors can affect a web page's performance. For this reason, truly effective Web Performance Optimization starts with identifying the most significant perf bottlenecks of your site. This is usually done with tools likeDevTools, WebPagetest, PageSpeed Insights, etc.
Once you've identified a possible lead, and taken the time to refactor and optimize it, it's important to follow-up by properly validating and understanding the impact of your change. Getting this right will help you learn whether that's something you should race to implement across your site, or a best-practice that in your particular case amounts to a micro-optimization.
This type of analysis is not trivial because web performance data is typically noisy. You can reduce noise by running your optimizations as A/B experiments side-by-side with the existing implementation, and by visualizing your data with a suitable graph such has a histogram.
This post explores these techniques in-depth.