Apparently, it was never actually "published" on Cypherpunks mailing list, just announced there. (Technically it was published on my "home page", as we used to say.) The original announcement can be retrieved from http://cryptome.org/cpunks/cpunks-92-98.zip, which is an archive of the list from 1992-1998. There are also two papers in Google Scholar that cite the 1998 b-money article (without mentioning Cypherpunks):

http://www.hashcash.org/papers/hashcash.pdf
http://cs.uccs.edu/~cs591/securityEngineering/jikzi-cpw.pdf

Here's the Cypherpunks post/announcement:

From: Wei Dai <weidai@eskimo.com>
Date: Fri, 27 Nov 1998 08:07:43 +0800
To: cypherpunks@cyberpass.net
Subject: PipeNet 1.1 and b-money
Message-ID: <19981126153349.A12001@eskimo.com>
MIME-Version: 1.0
Content-Type: text/plain



I've discovered some attacks against the original PipeNet design. The new
protocol, PipeNet 1.1, should fix the weaknesses. PipeNet 1.1 uses layered
sequence numbers and MACs. This prevents a collusion between a receiver
and a subset of switches from tracing the caller by modifying or swaping
packets and then watching for garbage.

A description of PipeNet 1.1 is available at
http://www.eskimo.com/~weidai.

Also available there is a description of b-money, a new protocol for
monetary exchange and contract enforcement for pseudonyms.

Please direct all follow-up discussion of these protocols to cypherpunks.

The idea here is that most people will accept the arbitrator's judgment (remember that both parties have to agree to the choice of arbitrator), unless one of the parties can produce clear evidence that the arbitrator behaved unfairly. So if Starbucks cheated you, and the arbitrator agrees with that, then Starbucks can still adjust their balances so that they have more money and you have less, but then they can't expect to spend that money anywhere else.