Operating System Support for Persistent and Recoverable Computations (1996) [pdf]
www-systems.cs.st-andrews.ac.uk
www-systems.cs.st-andrews.ac.uk
I think it might also be rewarding to look into combined language and OS approaches. The reason is that there's lots of diversity in operating systems, libraries, ISA's, etc with most projects standardizing around a certain language and interface. I could see some extension that marks things as persistent while compilers for specific OS's handle the details. Would help portability.
On other end, some research should be done in hardware support. Acceleration of persistence options at the least. I particularly think a hardware/software combo will help with safety-critical embedded where the hardware constantly checkpoints key state with the OS facilitating that with structuring and directing its operation. Well, no need to speculate as that's what Tandem's NonStop did with custom MIPS processors. A little in academia, too. I just think it's worth another look as it might not add much cost to an existing, high-end SOC in embedded.
External pagers (custom VMM and page fault handling policy on the per-process, per-task, per-library, etc. level) have been an integral part of all microkernel architectures since Mach, likely earlier on to the Accent and TRIX kernels.
The paper actually cited a reference to an old microarchitecture with capability-based addressing called MONADS-PC. Unfortunately, it seems like the paper is nowhere to be found in public, even though it's cited quite a bit! Instead, there's a shorter and related paper called "The Micro-Architecture of a Capability-Based Computer", but it's not the same one.
Far as MONADS-PC, that is interesting and might be one of those situations we were talking about with IT gems being locked up. I'll see if I can dig it out.
http://messagelab.monash.edu.au/Publications/DavidAbramson
Note: Acaddmics and research groups often squirrel away copies on their own sites for self promotion and sometimes public benefit.
Summary of the group's work below along with important point that hardware wasn't good enough for software at that point:
http://www.monads-security.org
Enjoy. :)