You can do the simple thing and use a pure silver heat spreader to transfer the heat to a room temperature superconductor that then is attached to a cubic metre copper fin heatsink.
That would actually do worse. More than a few watts come off the front (plastic) of the package, and the leads (especially the drain lead, but also the source if it's heavily bonded, which should be the case with those big-amp parts).
Otherwise that is just marketing BS and the real rating is going to be somewhere around 50W in real world apps.
Agreed. Under normal mounting conditions (finite heatsink, elevated ambient, insulating pads, etc.), 50W is usually pushing it.
Almost all the thermal resistance causing that limit is external to the part -- only in very small devices (e.g., IRF540 and down?) will the device limitations become apparent, and the power dissipation has to be derated further (30W, 20W...).
Or for lower junction temperatures, like 7805 regulators, which have limited die size to begin with, but a fairly conservative thermal limit built in as well. They're often put on paper-thin (not really) tabbed TO-220s, because they're only good for 10W to begin with.
They get that from the peak current multipled by the saturation voltage at that peak current. Keep it up for even 10mS and the device will turn into a small bright ball of plasma.
No, peak current is generally well into the desaturated region (assuming by "saturation" you mean the conventional sense, "Rds(on) limited region", and not "FET [current] saturation", which alas, is a thing, as confusing as that sounds), at which point the power dissipation is massive: 10 microseconds to failure, not milliseconds.
This may be slightly different for those really massive ones, the kind with lower Rds(on) in the die than in the bondwires. Though the outcome is still the same.
Tim