Load average based limiters have been in sendmail since the beginning; the
conf file options are

Ox - "load average at and above which sendmail will just queue mail,
instead of doing delivery."
OX - "load average at and above which sendmail will withdraw SMTP service."

NetBSD's default configuration file has them set to 8 and 12, respectively.
One could argue that those values should be reversed.

However, system load average reacts slowly to changes in load, and
particularly with the Internet being several million hosts, I believe it is
much better to precisely limit the number of processes doing SMTP service,
against the possibility that "too many" peers (for some value of "too
many") decide to contact my server at the same time. For a story about what
that situation is like, see "The UNIX Hater's Handbook", starting on page
85. There's also a copy of that story in the RISKS digest archives, from
May 1991.

The queue runners are a similar problem; if you have a load limiter in
there that will stop sendmail from doing work when the load goes above a
certain level, then having things set up to keep on spawning queue runners
is just asking for the system to get "stuck" at a high load, and not doing
any useful work.

Erik Fair