+++ This bug was initially created as a clone of Bug #1300253 +++ The problem's still there, and it's not just NetBSD. I just ran it a bunch of times on a CentOS test machine. It passes about 80% of the time, but 20% is an intolerably high spurious-failure rate. Marking bad on CentOS as well.
The job (netbsd regression) is not running anymore, and not planning to support complete regression on NetBSD for now.