I was following right along until your last sentence. I guess you're thinking about return in terms of getting back the money you put out for the card?
Yes - I didn't look farther out than about a year, so in that short window, the 2-cheaper-card setup would be just about as profitable as the 1 more-expensive-card setup.
My point was that two cards whose total purchase price was less than the fancier card produced the same output (about $7/day). So, if the output is the same, then the other factors to consider are initial purchase price, and cost to run. I ignored running costs in my conclusion, So let's look at that.
A 1060-6GB makes $3.70 per day and costs $0.29 per day to run (I put in my electricity rate, which is $0.12/KWH)
So, it stands to follow that 2 1060-6GB Cards will make $7.40 per day and cost $0.58 per day to run, leaving a profit of $6.82 per day.
A 1080TI makes $8.25 per day and costs $0.72 per day to run, leaving a profit of $7.53 per day
If a 1060-6GB costs $250, then 2 cards cost $500 divided by $6.82 = 73 days to pay for them.
a 1080TI costs $800 divided by $7.53 = 106 days to pay for itself.
So for a year, the dual 1060s make (365*6.82)-500 = $1,989
A single 1080TI makes (365*7.53) - 800 = $1,948
So if you stop after a year, the cheaper cards win. Obviously, the working lifetime of these cards is probably more than a year, so the more-expensive card will win at some point on the curve.
2 years would be (730*6.82)-500 = $4,479 compared to (730*7.53)-800 = $4,697. So there you are, if you run the rig a bit less than 2 years, then the 1080TI will be more profitable. Oh, hell - let's just solve the thing.
6.82X-500 = 7.53X-800
6.82X = 7.53X - 300
.71X = 300
X = 422.53
So as long as you run your rig longer than 422 days, it's better to purchase the 1080Ti than 2 of the 1060-6GBs.
Also counting against my initial conclusion is that it's pretty unlikely to gain a dual PCI-slot motherboard from customer castoffs, so that would mean more upfront costs on that side as well.
Then, you should probably factor in a certain percentage of downtime, equipment failures & such.
Yeah, I'm definitely looking at this from a hobbyist's perspective.