. . . at least with modules with a SIGNATURE we have some vague notion of accountability.

To add to what Abigail-II said, SIGNATURE files are not currently a good security system. As things stand now, it is unlikely that you have a sufficient web-of-trust to verify the author's key. It is thus very easy for man-in-the-middle attacks to work. Further, a lot of people don't check the signature until the automatic installation method has already done it for them (usually via a 001_signature.t test). This means the code has already started running by the time the signature is checked.

Right now, SIGNATURE files are only good as replacements for MD5 digests for verifying that the distribution is intact. IMHO, MD5 was already a perfectly good way to verify distribution integrity, so I don't see an argument for using SIGNATURE instead. (MD5 has its problems for use in cryptographic applications, but that doesn't mean it can't be used for integrity checks. If you're still worried, SHA1 can be used instead, which should still be faster than signature checking.)