> by longstanding convention, that port 23/tcp is given to telnetd type login servers

First thing I ever telnetted to was Melvyl, University of California's library catalogue, around 1985. This was “remote user access” (I was a remote user) to “service hosts” (running the catalogue) providing “remote terminal access”. It was not a login.

I remember using MELVYL too, and you're completely right about that.

I would suggest that MELVYL on port 23/tcp was also unnecessarily impinging on the IETF standards. MELVYL could have easily established its own well-known port with the IANA and not conflicted with the TELNET login port.

Before the WWW, there were a multiplicity of search services and indexes. Remember Archie, WAIS, and Gopher? Apparently, WAIS was assigned port 210/tcp, but Archie apparently used TELNET on 23/tcp as well.

I think some of the pioneering Internet services were perceived as not requiring a dedicated port. If MELVYL was the only service running on the mainframe and it wasn't running a Unix telnetd, then why not usurp 23/tcp? The admins there probably perceived it as a virtual "octopus cable" connecting remote "terminal labs", and for sure they had alternate methods of access for OS servicing and configuration purposes. In the beginning of MELVYL they were undecided about which protocol would prevail, and TCP/IP was competing with others, so port numbers may have been afterthoughts for the architects.

The most important thing may have been the principle of not surprising users or confusing them with parameters. "telnetting to a host" was way easier without trying to specify that they needed a port number. Just ask any Unix admin where MUD users try to bang on their telnetd port trying to play the game...