The value argument doesn't make sense if my original vision was a 2D pixel game that I could personally do in my spare time compared to being given the hottest 3D engine and now asset creation and dev is 10x the time and cost.
The big winners here look to be mid and big shops. Not to mention the asset stores these engine companies run. Having tons of indie devs suddenly buying all these 3D assets must be nice considering whatever cut Unity or Unreal are getting.
Terraria was written in .NET using XNA and met with huge success. But they too have seen similar problems and have to use external dev houses to get the game onto other machines.
Starbound is custom built. They messed up so badly with their initial build that it took them almost a year between releases to even release an update.
Kerbal Space Program _does_ use Unity, but even they hit a wall of shame caused by using the engine in a 32-bit variety. Mods are much easier because the platform is written using a "standard" framework, but the game will implode if you require more memory than 32-bit can address.
There are also plenty of success stories (Prison Architect might be my favorite game to watch develop) but I think the more complex the basis for the game, the smarter it might be to start with something that's proven and work your way up from there.
There are certainly games (e.g. visual novels) in which this wouldn't be very cost effective, but in most cases I believe that represents a huge value.