Bundler VS gemstash

bundle install uses the bundle binary (technically a ruby script). The command runs the Bundler::Installer.run module. There are a lot more steps, particularly around dependencies. It ultimately calls Bundler::Installer::GemInstaller.

sudo is totally unacceptable on my system

On the license in 2011: Bundler, which most people use to generate gem scaffolding, added the MIT license as default in 2011. I bet you'll see a change around that time where older gems are more likely to have a different license (Apache 2.0 maybe?) and newer gems are more often MIT. I'd also guess older gems are more likely to be indirect dependencies since they're probably (?) more likely to implement lower level pieces which were since wrapped up, so this could explain that phenomenon too.

If you are using Bundler version 1.x, you may also need to set the specific_platform configuration setting.

Sorry to hear. I'd say, never "bundle install" in the init script. If using docker, make "bundle install" an intermediate container. If you have your own servers, try network FS sharing your deps (and see if you're ok with the "require" latencies). Alternatively, use a gem cache proxy. You can use "geminabox" with a rubygems mirror, if you can tolerate the space and long mirroring times, or use a purposed "middleman" proxy, like gemstash, or nexus if your company is multilang.