I think the main point from a candidate perspective is that long, unusual or difficult interview processes are fine as long as we have already decided we really want to work for you (that is you are Trello or SpaceX or some amazing startup we love)
otherwise, just as you are faced with picking a rose from a faceless mass of candidates so are we faced with picking a decent place to work from a faceless mass of mission statements and "100%code coverage" blog posts (if we can find any)
And 2) yes I can believe it - that's what fizzbuzz and its ilk are for. The number of companies offering hackerrank/codility tests and justifying it as we only hire the very best has left the lake woebegone test way behind.
There has to be some
middle ground - where someone can pass a basic capability test to weed out the crap without demanding several hours of coding. I mean I spent two evenings recently polishing up a "Oyster card system for TFL" as a hurdle for a government department I have been angling to get in for a year. A no-name corporations wanting CRUD development can take a running jump if it thinks my evenings come cheap.
(amazingly at the turn of the century I used to use "has listed Python on CV" as a marker of "worth an interview" - it was rare enough and indicated an interest outside of corporate mainstream. Maybe Erlang today??)
Edit:
So my compromise suggestion is for companies to say "We have an OSS project (or even recommend one of the bigger ones). Take any of these 5 bugs and supply a pull-request"
At least then the evenings work is not totally wasted and you get to see an important skill - how fast can they come up to speed on an unfamiliar codebase?