And we could get into swap, etc. No, it's not airtight, and I know that. I'll just say that the archival feature goes beyond this level of leakage. Whereas most, if not all, of what you mention is at least contained in a single partition and in swap, and thus can be protected by full-disk encryption, creating a new hidden partition to store secret backups is violating expectations quite badly.
I choose to use a secure directory rather than full-disk as part of a security/performance tradeoff, and so I know that there are temp files and caches and swaps all over the place (though GnuPG is setuid so that it can lock pages in memory, thus preventing them from going to swap-- an attacker would have to freeze my RAM to get to my keyring). Most of these, though, should at least be manageable. I can see them and interact with them.
I do agree that in some sense TrueCrypt isn't doing its job, but I argue that's because their expectations have also been violated. Do the API specs make it clear that data written using those APIs may be copied off-partition without user interaction? If not, then the TC team would have to find out the hard way, then scramble to workaround this poor design decision.
GPG and PGP both make tempfiles, as do many of the email integration systems that use PGP, but I'm not here to help harden your idiosyncratic Linux setup, only to explain that your expectation that the OS designers are going to make 3rd-party crypto packages a priority is unrealistic.
If you need crypto-level assurance for your machine, you use full-disk encryption --- or at the bare minimum you turn off system restore points and use secure deletion software. People who harden Win64 professionally know to do this stuff, just like people who harden Linux setups professionally know the rest of the problems with your EncFS system.
I'm a security person, and not a Windows user, and I prefer the Win7 approach over the "whatever makes TrueCrypt easier to write" approach.
I choose to use a secure directory rather than full-disk as part of a security/performance tradeoff, and so I know that there are temp files and caches and swaps all over the place (though GnuPG is setuid so that it can lock pages in memory, thus preventing them from going to swap-- an attacker would have to freeze my RAM to get to my keyring). Most of these, though, should at least be manageable. I can see them and interact with them.
I do agree that in some sense TrueCrypt isn't doing its job, but I argue that's because their expectations have also been violated. Do the API specs make it clear that data written using those APIs may be copied off-partition without user interaction? If not, then the TC team would have to find out the hard way, then scramble to workaround this poor design decision.