Bugfix: we were checking to see if you want to send a keepalive, based
on five minutes since last successful write. But if you have bytes queued
already, and they're not getting through, we were adding a new keepalive
every second. This was bad.
svn:r4296