For me, it's the latency. If I'm researching something, I want to read the papers right then so that I can build off of them to do my next round of searches, instead of waiting for stuff to (maybe) trickle in. With scihub, I can do the search-read-search cycle in a matter of minutes rather than waiting hours/days for emails to be replied to.

May be field dependent but unfortunately the most likely outcome is the authors never getting back to you.

I have never heard about a researcher who would not provide an article when asked, ever. Most of them would also be happy to discuss the content of the articles, how they can be replicated and whether they are still accurate.

It does have more friction and latency than a search engine, though.