More problems self-hosting -current on RPi3

[bob prohaska]

Here's a strange buildworld failure with -current on an RPI3. 
 
To set the stage, a make -j2 buildworld/buildkernel was run from a 
clean (run make cleandir twice, rm -rf /usr/obj/usr/src) start. Kernel
and sources were at, or close to, r331146.
 
Buildworld finished, but the kernel build failed on an error of long standing,
something about mismatches between various versions of stdint.h in the file
system. Svnlite update reported quite a bit of activity, so the world/kernel 
build was restarted with -j2 and -DNO_CLEAN.
 
Some hours later, the system was found with all ssh sessions terminated
and the top output recorded in the link below. It looks as if the system
decided it was out of swap, and then went on a killing spree when lld.ld 
didn't go away. Disregarding the stated swap usage, ld.lld was less than 
1.4 GB in size with 3 GB of swap.

Details are at http://www.zefox.net/~fbsd/rpi3/crashes/20180330/

It's worth noting that gstat reports write speeds of 2-3 MB/sec to
the microSD card and USB flash, which does not seem much worse than
mechanical hard disks. The system _does_ seem prone to a sort of
"swap frenzy" when swap usage goes much over 10%, recovering at
intervals when all the cores are busy. Some months ago it was at
least occasionally possible to finish a -j4 buildworld, now even
-j2 buildworld is extremely uncertain. 
 
Is anybody able to self host -current in the last few weeks on an RPI3?

Thanks for reading,

bob prohaska