I'm fortunate that debugging those sorts of things (I'm more a systems guy for
a living) is something familiar; I'm accustomed to just chunking through
problems until it's sorted.

Turned out to only be about 72GB on disk once done. I think it took around
three days, but I had HTTrack running on a fairly fast VM in the same Linode
region as the webserver itself. This made things... snappy.

I did a lot of iterating through, then once I got everything seeming to work
fine I deleted the mirror, restarted it, and it ran completely through with
zero errors. After it finished I did a --update which took another 1.5 days
and had no errors, but that was more as a test.

Fortunately I was on vacation while this all was happening, so I'd mostly get
things going while having coffee in the morning, let it work during the day,
then check up at night. Worked quite well; I'm really pleased.

(Now I still have to decide if I really want to wholly move away from the web
app, because once in a while it's useful, but at least now I know it's
possible. And it answers some long-standing questions from the Gallery forums
about how to do this...)