One thing to take into consideration here is that when doing my migration work I had to add globals to the snapshotted state. I think for a successful execution the stack pointer (in global 0) at the end will be the same, but when migrating functions I had to add it to the snapshotted state to prevent memory corruption - the other mutable wasm structure is tables, but I'm pretty sure C compilers don't generate table mutations so it's not critical for snapshots to work properly.

Yes very good point; wherever we use the term "snapshot", we are currently referring only to a copy of the linear memory and not globals or the function table. As you say, the stack pointer should go back to zero after each execution, so ought not to be a problem in this instance where we're resetting after a completed execution. You're right that the table won't be changed under normal circumstances, but it will be modified when we perform dynamic linking. Migrating a function that has performed some dynamic linking would require adding the modified table to the snapshot, as well as loading the relevant dynamically linked modules on the target machine in advance of continuing execution.