>> The agp issue aside, I was able to track down where the crash
>> occurs (without the agp, of course): X crashes the machine
>> during init (xf86MapVidMem) in the call to i386_set_mttr from
>> NetBSDsetWC() in hw/xfree86/os-support/bsd/i386_video.c.
>
> there is a Option "NoMTRR" for the SCREEN section of XF86Config, maybe it
> could help?

For some reason I couldn't get the NoMTRR option to do anything
useful, X still tried to use the MTRR's with that option enabled.

But it does not matter because I already fixed the MTRR problem.
See port-amd64/37043 (I guess the PR has not been noticed, because
the patch has not been applied yet).