My best advice is: make measurements and be scientific about it. I regularly see weird claims and things that sound technically impossible. People often do things like make multiple changes and then misattribute the thing that actually helped, or make mistakes and misattribute the thing that is making it worse. Software systems are complex, results are not always what you'd think, and even experienced software engineers lose track of the things they've changed along the way.
If you think something is causing a performance problem, you should be able to measure it, make a single specific change, and then measure a significant improvement afterwards. If you're not doing that, it's not much better than rumour.