Keykos is very conventional in how applications are expected to organize and order their steps. The stack is replaced by a linked list of resume keys and the application can use these keys as it wishes to effect co-routines or one-shot continuations as it sees fit.
One calculation order that is difficult to arrange on most kernels is code that runs upon reference to memory of another program. Of course any plan available to a machine language program within one address space is possible.
Keykos developers had a number of ideas for physical distribution over multiple machines running Keykos but little was implemented. A critical issue is coordinating application logic with asynchronous checkpoints, or synchronizing checkpoints across machines. Neither is attractive. Other uses of geographic distribution may be uniquely possible with Keykos, such as one system writing checkpoints to a comm link instead of magnetic tape. Another distant machine continuously runs the checkpoint restart logic and is thus ready to instantly take over the load of the first machine should it fall into the ocean.