So are you arguing that random apache vulnerabilities are as likely to give simple, live, privileged memory access as they are to give access to the filesystem?? Assuming the system doesn't allow core dumps, this seems far-fetched.

In its conclusion the paper doesn't argue that developers should store passwords in plaintext in configuration files (which is the insane point you seem to be arguing for). It argues that passwords should be erased from memory when they're no longer needed.

Do you actually have anything useful to suggest?