With RAM sizes having ballooned to very large sizes (16-32GB now fairly common for a workstation) why is non-ECC memory even considered? Other methods used to safeguard large sums of 0s and 1s like hard drives, SSDs, modern filesystems (ZFS, Btrfs) have builtin error-correcting mechanisms. Why is getting hit with a cosmic ray and having a bit flipped in your CSV file any more acceptable than the same thing occurring on a "server" with ECC memory?
So, even if it weren't for the typical "enterprise/industrial" you'd be looking at a minimum of 12.5% parts cost increase.
Multiple that out * billions of ram sticks and you're talking real money for something with dubious relevance for most users.
Skylake-SP with ECC: $3,000
There's your answer.
Xeon E3-1230 v6 kaby lake 3.5 GHz - 3.9 GHz $250
Xeon E3-1240 v6 kaby lake 3.7 GHz - 4.1 GHz $272
Core i7-7700 kaby lake 3.6 GHz - 4.2 GHz $303
Xeon E3-1270 v6 kaby lake 3.8 GHz - 4.2 GHz $328
Core i7-7700k kaby lake 4.2 GHz - 4.5 GHz $339
I love competition. Though I would prefer if there was a third competitior in the x86 CPU space and the GPU space.