Someone benchmarked the two most-repeated Windows speed tricks. Neither does what the forums claim.
Open any list of Windows 11 performance tips and these two will be on it. Delete everything in C:\Windows\Prefetch. Turn off the SysMain service. Both sound plausible, both have been repeated for about twenty years, and Windows Latest put a stopwatch on them on Monday.
Prefetch first. The folder holds up to 1,024 small .pf files; on the tester's daily machine it was 364 of them taking 27MB. Each one records how often a program has run and the last eight times it started, and Windows only reads an app's file when you launch that app. Microsoft's own incident response guide describes the folder as "designed to speed up the subsequent launch of applications" - the clue is in the name. Emptying it on a healthy PC produced delays barely worth noticing. On a 4GB dual-core virtual machine, opening the Microsoft Store visibly dragged while Windows rebuilt what had just been thrown away. Microsoft developer Ryan Myers explained exactly this in 2005, and nothing since has made it wrong.
SysMain, which you may remember as Superfetch, keeps the programs you use most often warm in RAM. On a 16GB machine, switching it off changed nothing measurable in either direction. On the 4GB virtual machine it genuinely did speed up booting, 42.83 seconds down to 35.18, and then a cold ClipChamp launch went from 3.14 seconds to 8.48. You buy seven seconds once a day and pay five back every time you open something. Cached RAM is borrowed, not spent; Windows hands it straight back when an app asks for it.
Microsoft's own performance page runs to a dozen suggestions and neither of these is among them. It lists installing updates, clearing disk space, trimming what starts with Windows, and looking in Task Manager at what is actually eating the CPU. Dull, and that is rather the point. Tendvane's Health score covers the same unglamorous ground, which Windows build you are on, whether updates really finished, what loads at sign-in, because that is where the seconds genuinely are.