actually in-kernel telnetd (and perhaps rlogind) is in Solaris 2.4
and higher if I remember correctly, and the (ahem) express reason
was that at least one large customer was screaming about their
8-way Sparc1000 crawling under 200 or so telnet connections and
nothing else running on it... SunSolve bugreports are sometimes an
enlightening reading. BTW an all-time favorite is Adrian Cockcroft
saying "HP-UX can't have *this* POSIX-compliant _and_ running this
fast. Or if they have, somebody go and steal how they do it
!"... well, something very similar.
This is a classic SMP problem. Think about the things which UNIX
guarentees about the select() system call, now apply this to a
multiprocessor system. Wheee...