Why so many out-of-disk-space failures on buildfarm machines?

It seems like we see a remarkable number of occurrences of $subject.
For instance, right now we have these members failing on various
branches:

echidna No space left on device
asp No space left on device
herring No space left on device (icc seems particularly unable
to cope with this, or at least I suspect that's the
reason for some builds failing with that bizarre message)
kite gcc quoth "Internal compiler error: Segmentation fault"
wildebeest long-standing configuration error (no Tk installed)
wombat long-standing configuration error (no Tk installed)

I realize that a lot of these members are running on old underpowered
machines with not so much disk, but is it possible that the buildfarm
itself is leaking disk space? Not cleaning up log files for instance?

regards, tom lane