Just a few data points on the subject in support of the above.
My laptop (Intel 530 + Geforce GTX 950m) now works quite reliably with
in-tree X now. As the NVidia bit is not yet supported by our nouveau
port, I have "disable [nouveau|nouveaufb|genfb]" in my boot.cfg and
i915drmkms* configures fine. I can't quite figure out the
accelleration level - glxgears reports rather low numbers, however,
the xscreensaver demos work as expected. Firefox 52.9LTS also runs all
the WebGL demos I threw at it ( e.g. www.ro.me , rather nice). I tried
yesterday 2-3 times to build Firefox 62.0, so far without success;
will try later with a different option). Also when I run glxgears or
glxinfo, I get a message that this card instance is not supported by
libgl (or similar). This is when I startxfce4 from the command line
with /usr/X11R7/bin path before /usr/pkg/bin, as I had a few weeks
earlier compiled meta-pkgs/modular-xorg after having exactly the same
problem as the OP. Yesterday I also decided to try modular-xorg with
the i915 driver from wip as suggested by maya@, but the modular X
server crashed. The system is -current from yesterday.
All in all, -current is definitely going in the right direction, but
obviously needs work. I'd be happy to do any and every tests,
especially when there are news about nouveau and the expected later
transition to the drm2 from 4.8 kernel.
On a side note, -current (amd64) works fine under VirtualBox without
any modules compiled, only with an old build of VBoxClient from last
November - I just add 'VBoxClient -display' to the xfce4 startup
programs.
The samples are with xfce4, I've found it to be the easiest to setup
and best running of the full desktop environments under NetBSD;
otherwise I often run other less heavy display managers like evilwm or
i3. But if one is trying to get Firefox to run, one might as well go
the full mile and get the DE...
Btw I even tested ff52.9 with onshape, importing a few large stl files; works fine.