Its not feature creep that killed SC.
I'm a game dev and I followed its development just out of sheer morbid curiosity.. Pretty much their entire process was fucked from the beginning, and they never corrected it. Its a huge case of 'Escalation of commitment' rather than feature creep, being the root cause.
At the very beginning Chris Roberts hired some artists to make a visual demo in Cryengine, to raise money. It worked, great! .. This is when they should have scrapped the demo, and hired a really great lead dev to plan out the full game. But, no, they kept building on this amateur Cryengine demo in a disorganized fashion and have been ever since.
Uhm sorry but what you just wrote are pure assumptions. And the fact that you are "a game developer" doesn't make your claims more true.
If you really followed the development process and watched their "Around the Verse" video series you cannot, by any means, call it a "Cryengine demo".
In late 2013 the scope of the game got totally changed. It turned from a small multiplayer space flight game to an MMORPG. So they literally had to start from zero and nobody has any insights about the game development and organization except the CIG developers themself.
He's not entirely wrong though, CryEngine was and arguably still is wholly unsuited for anything that borders on MMO. Instead of realigning your plan around the changed scope, Chris Roberts decided to keep the glitter of CryEngine visuals and he is paying the price for this decision. Netcode is still affecting performance to this day and I doubt if it will ever be solvable.
We are long past the time when Iwata managed to single handedly refactor Earthbound in a few months with minimum external help. Complex projects carry so much inertia that makes it very hard to correct course once you have momentum in one (wrong) direction.
They switched 2 years ago from CryEngine to Lumberyard from Amazon which have an integrated feature for using AWS. They also stated that they completely rewrote the network part and still working on it to date.
They call it Object Container Streaming and Network Bind Culling (I'm not a network developer, so I have no idea about the technical side of it). I don't know how far along they are (they postponed the feature to the 3.3 update) but I'm confident in their abilites.
You do realise that Lumberyard is just an licensed build of CryEngine with AWS related APIs added on top, right? As of alpha 3.2 there is still a very noticable network latency correlated to the number of active players per server. I will gladly change my opinion when CIG do manage to fix it.
Still, none of this would help to explain why SQ42 was postponed to 2019 and beyond.
At the very beginning Chris Roberts hired some artists to make a visual demo in Cryengine, to raise money. It worked, great! .. This is when they should have scrapped the demo, and hired a really great lead dev to plan out the full game. But, no, they kept building on this amateur Cryengine demo in a disorganized fashion and have been ever since.