so i figured out that it is indeed worth doing at all right now and it's actually in a state of working (redeploying very large pexes can be ~instantaneous), without being super super ridiculous or requiring any breaking changes to pex whatsoever. i'm going to make a github issue detailing the changesets and their rationale as i break it into like 3 PRs for the separate parts (might have to be sequential PRs). i think this is a really generally useful changeset for all users of pex as it has alleviated an issue my friend who works on AR at facebook has told me about their use of pex for a while (regenerating massive binaries for their models takes a long time), and i think it can be introduced in a really gradual way that doesn't drastically change how people need to use pex to get these speedups. will post the issue link here and will post the first PR probably today. am extremely ready to accept review on these changes and will work hard to ensure they're mergeable and useful for the whole community.