I have encountered a number of instances now where ICs that supposedly include "thermal overload protection" will still heat up until they fry. Examples: LM338,LM2576.
These devices datasheets provide little details of the thermal limiting (temperature cutoff, reaction time, reset time, etc.) and also provide no power limit. Max power dissipation is simply stated "Internally Limited".
I no longer expect "thermal overload protection" to function at all unless the datasheet has a whole lot more data than the above.
Does anybody have any experience with these or other similar switching and linear regulator ICs actually shutting down non-destructively at a temperature limit, and is there some way I can roughly predict when this happens?