There is a way to extend the rewind limit with this method, too, that I didn't have time to implement: packing older chunks of state into the form of base state + replay, and unpacking them on-demand on a second thread. This way, the rewind period can easily be extended to the entire stage and more, all while requiring less memory footprint than the current 10s (actually 20s internally) implementation, at the cost of some CPU time.
I decided that it isn't really worth it to implement for the jam, though. I think that was a good decision.
I'm not sure if I'll be making a update to RSP this soon after the jam, especially with my LD48 game in need of an update too, but I might come back to it, some day, depending on the interest.
Thanks!