IMHO the question is one that no-one is likely to be able to provide a proven answer to, as the only 'fair' test would seem to require a lab setup in which a 'clean' Windows installation was maintained on the same hardware but never actually used for any other purpose than monitoring the OS use of resources. The install would have to be monitored throughout the whole supported life of the OS, receiving nothing more that automated updates and running nothing other than the built-in system maintenance tasks. The use of the built-in Browser and/or other built in software apps would likely significantly corrupt the validity of any 'test results'. So at the end of the day it's near impossible for users to come to any provable conclusion, as people use computers and their interaction with their computer significantly affects how the OS behaves and what level of resource it needs to manage user demands; i.e. the greatest impact on resource needs (RAM, CPU, GPU etc.) is 'you' and the software you want to use, not the OS.
Personal use (real-life) observations:
The nearest I have to an 'unadulterated' base system is a Windows 10 virtual machine that has existed in its present 'clean' form since the first official release (clean install over a former Insider Build) and has received nothing more than regular and feature updates when "switched on", but has never had any additional software installed; and being a virtual machine it has no 'real' hardware attached. 'Stress' is only really noticeable when Windows Update is active, but is no more apparent than on most hardware setups I've used and is mostly reflected in CPU demands rather than RAM. From personal observations only, this 'stress' is no greater than it was on day-one; in fact day one was probably worse because of the various automated setup tasks Windows self-performs post installation. All I can really add after the years it has been maintained it is that it now behaves exactly as it did when originally set up, with a minimal allocation of 2CPUs (=dual core processor) and 2GB RAM from the host machine. However, I also have working Windows 10 VMs with exactly the same resource set-up, but with a variety of software installations which still run comfortably with multiple applications active at the same time and they don't behave noticeably differently in normal use to the 'clean' setup, i.e. CPU & RAM usage only noticeably peaks during 'intense' Windows Update activity. Minimal 'crud' on these VMs as all user data is saved to drive space shared by the host machine, and Windows is 'cleaned' of accumulated Windows 'crud' at least once per month using Windows own Disk Cleaner.