In earlier work, we described a "pathological" example of a signature scheme that is secure in the Random Oracle Model, but for which no secure implementation exists. For that example, however, it was crucial that the scheme is able to sign "long messages" (i.e., messages whose length is not a-priori bounded). This left open the possibility that the Random Oracle Methodology is sound with respect to signature schemes that sign only "short" messages (i.e., messages of a-priori bounded length, smaller than the length of the keys in use), and are "memoryless" (i.e., the only thing kept between different signature generations is the initial signing-key). In this work, we extend our negative result to address such signature schemes. A key ingredient in our proof is a new type of interactive proof systems, which may be of independent interest.