The clang segfaults have returned and persist even after I got
rid of all the USB swap devices.

In a trouble report at
https://github.com/android-ndk/ndk/issues/386
a persistent clang crash is observed when cross-compiling
from (I think) X86 to aarch64. The report is old and
the problem supposedly fixed, but a workaround is mentioned,
using -O2 as a CFLAG. So, I added the line CFLAGS=-O2 to
/etc/make.conf after the second segfault building world.
This is evidently considered bad practice, but nothing better
came to hand.

It didn't work on the first try, there was another subsequent crash after
the restart) but after three repetitions of the build/install cycle the
clang crashes seem to have faded away. At r344015 svnlite reported

U    contrib/llvm/lib/Target/X86/X86ISelLowering.cpp
U    contrib/llvm/lib/Target/X86/X86InstrFPStack.td
U    contrib/llvm/lib/Target/X86/X86RegisterInfo.cpp
U    contrib/llvm/lib/Target/X86/X86RegisterInfo.td

It's unclear whether this bears any relation to the problem.

An update to r344113 successfully completed a -j4 buildworld/kernel and
install sequence without errors using CFLAGS=-O2, on the first try.

so the next test is
to try rebuilding chromium. A make clean was run in www/chromium, but
make re-used the old distfiles, so clearly nothing has changed in the
sources. Swap consists of 
/dev/mmcsd0s2b    4404252    45764  4358488     1%
/dev/da0p5        2097152    46204  2050948     2%
/dev/da1p2        3145728    46500  3099228     1%
at the moment. da0 is a 64GB Sandisk SDCZ80 usb flash drive,
da1 is an old 32 GB hard disk in a USB case. Both are connected via
an Amazon Essentials USB hub with separate power supply.