bad.js

var http = require('http');
var cache = {};

function curl(url, cb) {
  if (cache[url])
    // WRONG: synchronous callback
    return cb(null, cache[url]);
    
  var data = '';
  
  http.get(url, function(res) {
    res.setEncoding('utf8');
    
    res.on('data', function(s) {
      data += s;
    });
    
    res.on('end', function() {
      cache[url] = data;
      cb(null, data);
    });
    
    // WRONG: are you sure that 'end' and 'error' are mutually exclusive?
    res.on('error', function(err) {
      cb(err);
    });
  });

  // WRONG: the request object may emit 'error', but there is no
  // handler for it.
}

// Usage:
curl('http://www.google.com', console.log);

zoneified.js

var Zone = require('./').Zone;
var http = require('http');

var cache = {};

function curl(url, cb) {
  new Zone(function CurlZone() {
    if (cache[url])
      return zone.return(cache[url]);

    var data = '';

    var req = http.get(url, function(res) {
      res.setEncoding('utf8');

      res.on('data', function(s) {
        data += s;
      });

      res.on('end', function() {
        cache[url] = data;
        zone.return(data);
      });
    });
  }).setCallback(cb);
}

// Usage:
curl('http://www.google.com/', console.log);
curl('http://does/not/exist/', console.log);

I agree it's a bit vague. The key is this line:

curl('http://does/not/exist/', console.log);

Because the async op is wrapped in a zone, the error is passed to the callback as the first argument, which in the example happens to be console.log.

So what you're seeing is basically the output of:

console.log(error);