As I have mentioned in other posts, it has happened to me several times. The difference is that my game, being simple in terms of resource consumption, is not so evident (which made me think that it was a profiler problem and not Gdevelop’s).
I suggest you run short performance tests, as short as you can. I have realized that the consumption of many events or sections is diluted over time, giving an underestimated value about the real impact while they are executed. Events that are executed 1 time every second for example, you will be diluting their impact in another 59 frames in which they are not executed (which gives a lower total value). With this I am not saying that the performance is random due to the events (it is not logical that with the same events, optimized or not, it works completely differently), but it will help you investigate more specifically. I’ve seen within the rendered section that sometimes (with tests of about 12 frames), the value is almost doubled, without anything different or added to justify it. I figured there might be a problem with the profiler, but I don’t know, maybe not. So I suggested that testing could be done by selecting ranges of frames (Profiler in the debugger with frame by frame resource consumption analysis).