IMO Bitmain has done themselves a disservice by trying to build a 1TH miner with their existing ASICS.
As an EE and embedded software engineer I see these problems:
1. Bus based systems using card edge connectors - unreliable (see original IBM PC)
2. Beagle board is hobbyist grade - unreliable, known for SD card problems
3. Too many ASICS - unreliable
4. Shipping problems because of #1 above.
The S1 is rock solid because it was designed that way.
1. Two blades solidly attached to a metal frame and connected with ribbon cables.
2. Commercial quality controller in the TP-link
I assume (hope) Bitmain cobbled this together as a stop-gap while they ready their next gen ASICs.
Bitmain responded to customer demand. Quite a number of people had issues trying to colo the S1 because of its form factor and expressed a strong desire for a contained 4U 1TH unit. They (Bitmain) delivered.
As per your other points:
1. I agree that there are and were issues with card edge connetors, but the industry still uses them at large. Manufacturing and electrical component characteristics have come a long way since the 70's and 80's.
2. I agree.
3. Greater number of chips allows for redundency. You can have a single ASIC fail in an S2 and be okay, if you have a Cointerra or HF unit and a chip fails, that's a large percentage of the machines achievable hashrate and would warrant an RMA. In the world of Bitcoin mining an RMA is a death knell.
4. The shipping issues aren't a result of the card edge connectors, it's a result of deficiencies in the mechanical design and packaging. Computers ship all over the world everyday with card edge connectors in place and arrive just fine. IMO there was inadequate support for the guide rails, poor clearances (loose) between card edges and guides, and no mechanical lock securing each card.
