If direct (non proxied) access from the search engine spiders can be identified serve the real robots.txt. Otherwise disable crawling. Also, switch meta noindex like this.