Most likely it is so that sendmail can determine the load average. It
has a feature that it will start queueing everything (rather than
delivering immediately) at one load threshhold, and another threshhold
at which it will begin refusing connections altogether. On some
systems, there is a system call to get the load average through the
kernel. On those where there is not, it is necessary to read /dev/kmem
to get this information.