I cannot say I was ever satisfied in in every context where I saw this model applied. My experience ranges from trivial languages such as this, to compositing software, DAWs or something more sophisticated such as programming-oriented signal generators (where blocks are often used as filters/generators and edges as I/O connectors), to workflow editors...
It "might" look intuitive at first[1], but the network just becomes unreadable as soon as you have more than 10 nodes or non-linear behavior (with edges eventually crossing). Auto-layout just becomes a major problem to be solved. Auto-layout also destroys the visual locality of your program, so it actually goes against the programmer for anything non-trivial. Manual layout is something I do not want to take care of either.
Is any experiment or research being done that goes in a _different_ direction?
[1] I would actually say "less scary". I'm not convinced a simple programming language in the spirit of "BASIC" is less difficult to comprehend in the technical sense.