Runaway forking on ElasticSearch servers

Hi -- I'm running ES 0.19.9 across 3 servers whose fork rate usually sits
under the 10 per-second mark. We've noticed for a while that these servers
will, at seemingly random intervals, suddenly jump to a fork rate of
between 90 & 140 per-second, and having jumped up, there they'll stay until
ES is restarted.
Could anyone advise me what might be causing this in ES, if I should worry
about it, and what maybe I can do to stop it happening?
Thanks very much,
Doug.

--