The genius of SpaceX's approach is that they built a landing system where it's OK to be unreliable for a while. Instead of being a billion-dollar item like the Space Shuttles were, the Falcon 9 boosters are a few tens of millions of dollars apiece, and the launches are priced to cover the entire cost of them plus more.
ASDS repair depends on the nature of the failure. When a low-margin GTO mission miscalculates and craters, that does indeed take some fixing. A JASON-style tip over and kaboom seems to basically involve picking up the pieces and repainting the lines.
I think your last sentence reverses cause and effect. If they substantially cut down on core construction, that will be because the landings became reliable enough for that to happen. That doesn't make failures costly now.
I will also note that their landing approach is inherently unreliable to a degree, because there's little redundancy. If Falcon 9 loses an engine on the way up, it's fine. If it loses an engine during the landing burn, it's going to kerplode. If it loses an engine on the way up, then that will likely wreck the fuel margin needed for landing. I'd be shocked if Falcon 9 can ever achieve anything like the Shuttle's perfect 133 out of 133 landing record. (I'm not counting Columbia since it was ultimately a launch accident that manifested on reentry, and never got anywhere near the actual landing.) But that's fine, they don't need to.
For booster landings it's probably okay to be unreliable forever as long as the failure rate is estimated properly and the value of recovered boosters exceeds the cost of developing and maintaining recovery capability. Since there are no people or cargo involved it's a straight exercise in accounting.
In cases where they are using three engines during the landing burn, losing one would be catastrophic. You'd lose a third of the thrust you were counting on, and the remaining thrust would be highly asymmetric. The other outboard engine can't gimbal far enough to keep the thrust vector through the center of mass of the rocket.
Plus within the next year they are basically going to stop producing new cores, so loosing one won't be "cheap"