Temperature variation is the real card killer. I'd agree with anything < 80-85C, but some cards just don't stay that cool even standing alone with good ventilation and room-temperature ambient environments.
It's more important to make sure your temperatures aren't swaying significantly on a regular basis, the (small) expansion/contraction will eventually stress the metals in your GPU, leading to reduced life.
A stable 80C is far better than swinging between 60C and 80C every day.
This isnt quite true. Heat cycling is a small risk for haircracks in solder and underfill. This problem has become well known when a batch of nVidia cards suffered from a manufacturing deficiency that made them extremely vulnerable to this. These are the cards you can usually "fix"by reflowing them in an oven. This is something that pretty much never happens with AMD cards.
High temps otoh increase electromigration. Electro migration can be thought of as "internal wear" inside the chip. Its what over time leads to lower overclocks (if transistors in the critical path are affected) or instability and plain old dead cards. This is far more frequent, particularly with mining cards. Electromigration is a function of heat and the relationship is exponential.