Assuming first GET will succeed and return quickly in 99.9..9% of the cases, so always sending a second GET seems wasteful, as it would put constant additional load on the network, CPUs and disks.
Here's how I'm thinking about that; the costs of network transmission, computation and storage are all decreasing at a greater rate than the global economy (and population) is growing. Per person (or spending-unit) they are getting very cheap.
At some point, those cost-reduction curves plateau and the improvements become marginal - but at whatever rate they settle, will the costs be so low that using many systems as a form of massively-distributed optimistic concurrency will be a no-brainer?
And at the meta-level; are the benefits for human interaction compelling enough to themselves help drive the costs down.
