And, to follow up on my own posting (sigh), RFC 3235 and 3027 are Informational... we have no STD, and no BCP, that come up when you search for NAT or Network Address Translator, so... perhaps there is no community consensus document that says what the community consensus appears to be, and the best thing to do is to Google "NAT end-to-end" and leave the result as an exercise for the reader?
There was a discussion on IETF ML in April and May of 2000 on
draft-ietf-nat-protocol-complications-02.txt
and, vint, for example, wrote:
Date: Sun, 30 Apr 2000 06:20:48 -0400 From: "vinton g. cerf" <vcerf@xxxxxxx> Subject: Re: draft-ietf-nat-protocol-complications-02.txt
that's right - they use iMODE on the DOCOMO mobiles. iMODE and
WAP seem to have that in common: a non-IP radio link protocol
and an application gateway. Of course, this limits the applications
to those that can be "translated" in the gateway, while an end to
end system (such as the Ricochet from Metricom) would allow essentially any application on an Internet server to interact
directly with the mobile device because the gateway would merely
be an IP level device, possibly with NAT functionality.
So, according to vint, NAT is less evil than gateway translation.
Of course, the technical reality not tainted by commercialism is that NAT translate protocols at the IP, transport and application layers.
But, you can't expect much help against NAT from IETF.
Masataka Ohta