Another idea:
These are RF MOSFETs, which look like a mix of an SO-10 package with a bad case of webbed feet, and a D2PAK or something like that, for the size of body and heatsink tab. Or an SMC diode with a slightly wider body, much wider leads, and a huge heatsink pad.
The construction is like so:
First, carve a slot in the board. The heatsink tab is a hunk of 1-1/2 x 1/8" copper plate, about 8mm long (to match the heatsink pad on the devices). The hole is tight enough to provide a friction fit (yes, cutting and filing the hole was tedious -- I have no CNC, so I did it by hand). Tap it in place until the component side is level with the surface of the PCB, then solder support straps across (the extra hunks of copper clad straddling the top and bottom ends of the bar). This takes a lot of heat, but mainly it takes a lot of time (and a heat resistant surface) waiting for the copper bar to heat up.
When the plate and devices are soldered in place on top (also, a piece of copper foil, slathered in solder, shorts across the top side copper), go around the bottom, soldering a fillet. This both shorts the plate to the bottom side ground, and provides additional mechanical mounting.
Also, before the copper plate goes in, take the time to cut at least a few traces. I carved out the 6mm wide (10mm length, 1mm gap) pads for the transistors (gate and drain, respectively) before soldering in the plate.
This is going to be a ~20W push-pull RF amplifier, which should dissipate about 20W at ratings. I'm still designing the transformers for source/load matching. Taking it methodically, since it should be capable of ~500MHz bandwidth. (The transistors themselves roll off somewhere in the GHz!)
Tim