This is a great point that's often overlooked. In fact, this is why I don't do full system / state backups of the client PC's on our network anymore. Restoring those type of backups seemed to create much more of a headache than restoring basic file & folder backups. I don't try to capture every little aspect of a persons profile, screensaver, display resolution, browser history, etc. I point my backups to only what I know I absolutely have to have, which usually has nothing to do with windows system files. 90% of the time, backing up a users My Documents and an e-mail store (.pst usually) is all I need. We used to run full nuts to bolts system state backups and store them on tape drives, Yosemite I think. Restoring from those types of backups never once went as straight forward as it should have, and it was expensive to boot. Since having changed to only backing up user folders, I've never once had an instance where I wished I had a full system state backup to work with. When only dealing with files and folders I have a lot more options for usable backup media as well. YMMV.
Best way to go when you KNOW what is in the environment, yes. ISTR WinWOES had abut 70,000 "drivers" in .cab files alone before I ceased to care. No backup needed, faster and safer malware scans if they - and about 2/3'rds of the other trash Win installs with itself - were REMOVED from an active machine once configured.
Did a few where "RAID 1" mirror sets had three elements, rather than two.
Trained the gal clerk/receptionist to rotate the pull-out SCSI trays last thing each day before locking-up, set the "Friday" and "Friday of Fridays" aside. Scripts performing a continuous detection exercise did the rest. Separate system checked their health wee-hours, could page more expert hands to the site before the next workday opened.
Fast-forward ten-plus years, had Matt Dillon and the then-current code maintainer make a minor tweak to 'cpdup' source code, ran it out the backside, rack-side, internal use NIC amongst RAID 1 servers over 10GigE on IPV6. Cycled at an odd number of seconds from each end to reduce risk of collisions or costly overlaps.
Large, but rapid set of hand-shaking transactions - really lean as to data, per-each - then seldom more than a handful of files to actually copy, run at sub ten-second intervals as it was.
Slower clone ran an off-continent 'cpdup' at a lower data rate, stayed about one-full day back, worst case.
No actual 'restore' operation visible to end users in either case.
Simple replacement of a failed mirror-set member or server, entire, then background automagical rebuilds.