weird buildfarm failures on arm/mipsel and --with-tcl

one of my new buildfarm boxes (an Debian/Etch based ARM box) is
sometimes failing to stop the database during the regression tests:

http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=quagga&dt=2007-01-08%2003:03:03

this only seems to happen sometimes and only if --with-tcl is enabled on
quagga.

lionfish (my mipsel box) is able to trigger that on every build if I
enable --with-tcl but it is nearly impossible to debug it there because
of the low amount of memory and diskspace it has. (two consecutive
failures will run the kernel out of memory due to the resources consumed
by the still running processes).

After the stopdb failure we still have those processes running:

pgbuild 3389 0.0 1.5 39632 4112 ? S 06:14 0:03
/home/pgbuild/pgbuildfarm/HEAD/inst/bin/postgres -D data
pgbuild 3391 0.0 0.9 39632 2540 ? Ss 06:14 0:00
postgres: writer process
pgbuild 3392 0.0 0.5 11220 1348 ? Ss 06:14 0:00
postgres: stats collector process
pgbuild 3488 0.0 2.4 43640 6300 ? Ss 06:15 0:01
postgres: pgbuild pl_regression [local] idle
pgbuild 3489 0.0 0.0 0 0 ? Z 06:15 0:00
[postgres] <defunct>

Any ideas on how to debug that any further ?

Stefan