Catalyst's current (2.0_rc46) handling of autoresume is a bit...dumb. I'm not saying that the design/implementation was done stupidly. It's just that catalyst is "dumb" about it.
For example, I had a messed up build at /var/tmp/catalyst2/tmp/default/stage3-i686-2006.1-pre/. I completely removed this directory. I neglected to remove the corresponding .autoresume directory for this build, so when I went to start the build again, catalyst died almost immediately saying that it could not execute /tmp/stage3-chroot.sh (or whatever it it's called).
Basically, for each check for a file under self.settings["autoresume_path"], there should be a corresponding check for a file that *should* exist in the build directory if we are actually ready to resume at that spot in the build. It appears that some of the autoresume_path checks have corresponding sanity checks, but I've been bitten by this particular one twice.