Back then I was more impressed by NT and OS/2. Windows 2000 is the one I remember most fondly.
95 dealt with every weakness of NT in those markets and provided developers, consumers and SMBs with a smooth migration path towards it. It had very good backwards (DOS and 3.11) and forwards (XP) compatibility, reasonable hardware requirements and the ability to boot into DOS (important for gaming and some other applications at the time). It also offered massive advances over 3.11 (proper multi-tasking, the best looking GUI of the time and greatly improved stability) that when combined with the above, pushed user adoption.
Given its design requirements, 95 could never have been as stable as NT, but it was a spectacularly successful interim solution for customers not yet ready for the trade-offs that would have involved.
—-EDIT:
I misremembered. It was NT 3.51 that still had the video driver in userspace, NT 4 moved it into the kernel: https://en.m.wikipedia.org/wiki/Windows_NT_4.0
For me, the key Windows 95 moment happened while I was sitting in an IBM lab working as a tester for OS/2 Lan Server. We had a copy of Windows 95 we were testing for compatibility.
Playing around one afternoon, we decided to try to break it. We started Drivespace compression, let it run for a while, and pulled the plug on the machine. Plugged it back in and it picked up right where it had left off...
I'm pretty sure it's possible for a DOS process running in a window to issue a file system request (a DOS API call) that gets trapped and processed by 32-bit VMM code and then immediately forwarded back to (a different instance of) DOS for handling by a legacy device driver.
There are reasons for this.
It's amazing it worked.
I'm thankful it's gone.
While Windows 95's kernel didn't have the full feature set of NT, it still was more sophisticated than DOS.