Satisfaction of level 1 of this requirement will require changes only to Pluto, specifically to provide a way to get a list of current connections, to record this in a stable file, and a for the boot up scripts to read the alternate list of configurations as well. So, this requirement can be satisfied without impact to KLIPS2 design.
Level 2 of this requirement has some issues. The storage of keying material on disk may be a source of concern. This issue would need to be addressed in the design. The source of this requirement is to provide reliable recovery and fast reboots, systems that involve operator intervention may not satisfy this. The chief advantage of storing the phase 1 information is that it reduces the amount of time required to do DH exponentiation after a reboot. A new phase 2 would have to be done as well.
Level 3 of this requirement has further issues. It requires some help from KLIPS2 to provide for the retrieval of keying materials (including replay state) from the kernel, and subsequent reloading of it. There are clearly even more issues with making sure that the materials are not inappropriately revealed. In addition, the state of eroutes, filtering, etc. will need to be captured. Saving of this information may have very strong advantages in the opportunistic case, as the information on whether or not to set up an opportunistic tunnels is valuable as well. Further, in the opportunistic case the risk of disclosure of the keying material may be considered low enough that storing it is worthwhile.
In all three cases, there is a cost-benefit analysis to do, weighing the improvements in reliability and performance against the risks of inappropriate disclosure. The answer to this analysis may always be a local matter.
In addition, all three cases would apply to restarting of Pluto either on purpose (to facilitate easy updates), or due to program error (core dump).
There are further legal issues. Access to the keying materials may facilitate cooperation with law enforcement access. This is not regarded as a feature.
Opportunistic encryption would benefit from any amount of key maintenance. Road warriors are the ones most likely to benefit as they are turned off/suspended most often. However, their wildside address is also most likely to change, rendering any saved state that they have useless.