|There's more than one way to do things|
Whilst I think that your discussion of the problems is correct, I think your choice of algorithm and your code for demonstrating the problems is problematic. The latter in part because you are using Perl, which I specifically excluded; in part because you then use Coro which IMO clouds the real issues and adds more of its own.
Firstly, Fibonacci is a poor example because it is simply so easy to achieve better performance through mechanisms other than threading. As the only purpose of threading in this context is improving performance, using them for Fibonacci is never going to work effectively.
Whilst the recursive Fibonacci algorithm doesn't lend itself to tail call elimination, it is easily written iteratively for an huge performance gain. Equally, the recursive algorithm can be sped up immensely by simply memoising it.
But more importantly, the iteration/recursion can easily be done away with entirely using Binet's formula:
All of these modifications make it possible to calculate Fibonacci( 1474 ) (the limit of a doubles ability to store the result), more quickly than it is possible to spawn a thread. At least using the current implementation of iThreads. And possibly Coro also.
I realise that you chose it only as a well-known example on which to base your discussion. But if you are going to explore the threading of recursion in Perl, there are simpler ways than your implementations. This does the job:
Where the futures module is:
Using three subs serves to highlight the level dependency, but these could be merged into one sub with conditional statements provided there was some way to determine the number of existing threads.
Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
"Science is about questioning the status quo. Questioning authority".
In the absence of evidence, opinion is indistinguishable from prejudice.