I'm currently on the hardfork. After some analysis and testing, I think that the best is to start with one superblock of (about one hour) per week and break the record for sextuplets. We would add an entry to the top ten once per week.
After that and with more time we can think of breaking the others. I've been thinking on how to work with much larger tuplets that would take hours or days to find, without the need for superblocks. It would work similar to how pooled mining works: blocks would allow tuplets with one or two composites until we finally find the record tuple of all primes. Doing this we could obliterate all records for tuples of size 6 and onwards!
Tuplets of fewer primes, like twin primes or even single primes are still out of reach until we have more ideas because block verification would take too long.
I was considering the idea of having two threads going on simultaneously with mining a "superblock" on one and the regular 2.5 min transactions blocks on the other. Possibly generating 1/2 of the coins on the superblock and 1/2 on the standard blocks. That way we don't gum up the transactions while processing superblocks.... That being said, I'm starting to dislike the idea of superblocks interrupting the 2.5 min transaction cycle less and if we want to do a hard fork we should do it one time only and get it right.