First, a quick overview on how the landing works. After the first stage separates, it executes a "boostback burn," using the engines to get it moving toward the landing site. Sometime later, it fires the engines again as part of the "reentry burn," which is used to soften the blow of hitting the atmosphere at Mach Whatever. Apparently the rocket exhaust acts a little bit like an ad hoc heat shield, plus it helps slow the thing down. Finally, right as it approaches the landing site, it fires the center engine a final time to, hopefully, bring the rocket to a stop at zero altitude.
The amount of fuel used varies a lot, because they can land in different places. At one extreme, you have a landing at the launch site, which uses the most fuel in order to get the first stage turned around and coming back. Barge landings use less fuel, because they can be positioned under the flight path so the rocket doesn't have to maneuver as much.
SES-9 was at the other extreme. It reserved the bare minimum of fuel for the landing attempt because of its heavy, high-performance payload. That meant no maneuvering boostback burn, no or minimal reentry burn, and the landing burn itself was done with three engines instead of the typical one. That saves fuel because the rocket decelerates faster and spends less time fighting gravity, but of course the control challenge is magnified, which is probably why it went all explodey.
Putting some numbers on it, I found some estimates here:
https://www.reddit.com/r/spacex/comments/2vuy1v/how_much_fue...
According to that, landing at the launch site would require about 38 tons of fuel, and a (normal) barge landing would require about 20 tons. SES-9's barge landing would have been even less.
I believe the 15% from your StackExchange link is confused, and someone mistook the payload penalty (about 15% for a barge landing) for the amount of fuel required.