I just encountered this problem and tracked it down - when syslogd
restarts, a bug in the syslog(3) code prevents the daemons who have
existing log sockets from discovering the new /var/run/log, and their
subsequent log messages never make it to syslogd (they do get written
to /dev/console in most cases, for what that's worth).

This was reported back in late September 2000 - it's fixed in current,
but it never seemed to have been pulled down into any other releases -
a bit of a drag, consider the potential for loss of log messages.

Also a search for "syslog" on the Netbsd site's query-pr page doesn't
turn up this (or any other problem), though the Google link did.  Guess
I'll use that next time I go bughunting.