Rule of thumb: more than 50W for a TO-220, 100W for a TO-247, etc., is getting into risky territory. It may be fine, or you may need specialty materials and methods (better insulators, heat spreaders, water cooling).
The Tc=25C power rating of the device, is measured under boiling Freon. When they say "Tc" (case temperature), they mean
the whole case! It is a meaningless figure.
Expect to spend about double the RthJC on an air-cooled heatsink, and double again on thermal interface material. This basically brings a meaningless "300W" figure down to a more sane 60W, hence the 50 and 100W ballparks.
Heat dissipation, and 2nd breakdown, track fairly well with respect to die size. Older FETs are advantageous for this reason: I tested (to destruction) an IRF740 at much higher power level than the datasheet suggests. (The gotcha is, they are rated for
maximum RthJC, but clearly, the
typical RthJC can be much lower. Can a measured RthJC be relied upon? Who knows. You better have a good QC program in place...) Meanwhile, I tested a new production 6N60 (I forget what exactly, FDPF6N60 or something), which failed at something like 105% of ratings, in other words, spot on.
Taking them apart, the IRF740 die was easily triple the area of the FDPF6N60.
2nd breakdown isn't an issue for low voltage parts, or at low voltages; for high voltages, it depends on type. Most modern, high density switching transistors are not suitable. Older transistors largely don't care: just make sure DC is on the SOA. The newest high voltage transistors (SuperJunction) apparently are not susceptible to 2nd breakdown, and are often seen with straight DC SOA curves.
As for "linear" transistors: yes, they're more expensive. So what? As kinds of parts go,
transistors are one of the cheapest components you put into any project. They're surely beaten out by resistors, but you'll easily spend more on capacitors, or transformers, in a typical project (given that I'm talking about projects a little more involved than a mere current sink). Particularly relevant here: the heatsinks, and all the mounting hardware, and all the labor to stick them together. Another five bucks for a good transistor is a stupidly small cost on top of hours of labor. Stay real about your priorities!
Tim