It's about the drivers. Getting traction with a new OS is all about driver support. Which is what makes launching a new OS hard, assuming your goal is to be popular. You have to inspire an army of driver writers.
But was that better driver support simply a snowballing consequence of Linux already having been ahead by then? If so (and IDK if it is so) then the original explanation would have to lie elsewhere.
There was quite a bit of weirdness surrounding the BSDs back then due to AT&T code that was replaced, it kind of became moribund after some dispute over patches that I didn't follow too closely, then those patches became forks (FreeBSD and NetBSD). By the time the smoke cleared, XFree86 was ported to Linux and the general perception was the Linux community had their stuff together, while the BSD guys were fighting over nothing.
I don't think it helped that the BSD distributions were very opinionated about their userlands, where in Linux you had lots of choice.
In hindsight, it seems to me the BSD guys simply blew their shot by infighting at a time when early adopters were clamouring for a Unix like, and didn't care about the BSD purity war.
That was where the snowball started. Linux had no real advantage other than the cooperative nature of the community.
Sounds like it's getting back to people then. It still seems like a strange puzzle to me.
But it seems plausible that someone, somehow, managing to get one important driver into Linux that was missing from BSD might have helped to get the ball rolling, enabling "Linux has better driver support" to become a self-fulfilling prophecy, encouraging driver writers to write drivers for Linux (rather than BSD) and for Linux to include them. Presumably BSD eventually largely caught up, but by then the race had already been lost.
It would definitely have been something like that. There weren't an infinite number of devices for PCs back then, but every one enabled more people to run Linux. Would be interesting to try to track it back to see where exactly interest really started taking off.
For me personally it was the Panasonic/Mitsumi interface CD-Rom on the Sound Blaster cards. Vague and possibly incorrect memory suggests 386BSD (and maybe FreeBSD/NetBSD) you could get an expensive SCSI cd-rom working, but not the Panasonic/Mitsumi ones that were vastly cheaper and more available.
Because the internet simply wasn't available in a form useful for downloading a distribution, CD-Roms were the only practical way to install a Unix-like. Cover discs, or Slackware, or Yggdrasil were available at your local book store.
The MPC[0] standard was kind of on it's knees during that period, so it was pretty hard to keep up with the variety of hardware on desks at the time.
This old usenet post from 1992 [1] has contemporary reports from where 386BSD was competitive, but the general consensus even then was "Linux works on more affordable hardware" so the problem must have existed before that post.
edit: [2] This salon article seems to point the finger at the Jolitz's approach to development.
The insanity of how they'd take Binutils/GCC and split it up like a BSD tool, and twist it unrecognizably was just so odd.
I'd been collecting old BSD stuff from the era, and even 4.3BSD/Mach is so weird as well, although Tahoe i386 was doomed for not having source available. A larger issue with the platform as it wasn't open and code wasn't contributed back.
It's about the drivers. Getting traction with a new OS is all about driver support. Which is what makes launching a new OS hard, assuming your goal is to be popular. You have to inspire an army of driver writers.