We have seen this when the radius server gets busy. The default is to do
a sequential search through the users or passwd file; when the file is
long and the load is heavy, the search can't be completed before the
Portmaster times out and sends a second request. This is why you're
getting the dropped requests and the "unresponsive child pid" messages.
For us, the workaround was to go to a dbm version of radiusd, and when we
couldn't get that to work, to write a hashed lookup of our own. Neither
of these works with the passwd file (at least, not so far as I know).
There may be a similar solution for searching the passwd file.
-- Bruce