New RLBox-generated code hits a slow path in valgrind, relating to memory
address state changes associated with sandbox heap creation/destruction. This
only happens when valgrind is instructed to track the origins of undefined
values, which is a non-default setting requested by the automation run. See
valgrind_args in build/valgrind/mach_commands.py.

A proper fix requires investigation/fixing in valgrind. For now, we can
perhaps work around the automation run timeouts by increasing the timeout,
perhaps from 1800 (as now) to 2400. Removing --track-origins=yes would also
"fix" it immediately, but would cause a loss of useful information in any
undefined-value error messages reported by V.

Further ahead, we should upgrade valgrind to latest trunk (3.19.GIT) + fix for
this problem, if one can be found. That can be a separate bug though.

https://hg.mozilla.org/mozilla-central/rev/c78d918c4111

The underlying bug, plus testcase, was filed upstream at
https://bugs.kde.org/show_bug.cgi?id=446103