Performance is certainly a concern, but there’s a bigger problem: how do you read a consistent state out of a spreadsheet where every cell is a separate process? What if some of the cells are doing a long-running operation? It’s possible, with additional machinery, to do this - but it winds up looking like a Paxos knockoff because it is.
I’d recommend focusing on the pure-function part you mention above (compute etc). Then those functions can either be composed together sequentially (iterate over a list of changes to produce new changes, repeat until fixed point) or concurrently (split into communicating processes), but that’s an implementation detail.
