My experience is on QCX+ rather than QDX but for what its worth :

Having put a very very light smear of heatsink compound (standard white silicone/zinc stuff) on the flat faces that get screwed to the PCB there was still quite a bit of heat in the top sides /steel washer on prolonged continuous 5W+ TX with the case closed. 

Yes, I know that nowhere does Hans suggest that prolonged 5W+ TX is what the units are designed for.  In fact the opposite 

As I'm interested in WSPR more than CW  the usual duty cycle saviours may not apply to my case either.  However Ham radio is experimentation and I have a drawer of BS170's handy..so

I drilled and bent a heatsink from a significant area of 1mm flat copper sheet to nicely fit the round upper surfaces of the transistors and carefully set up ultra-thin compound along the ridges and a thicker band either side of each package so most of the curved surface has compound between it and the heatsink.   The central 5mm x 1mm of each transistor case is in full contact and will have a reasonably good thermal path  to the heatsink, The same again on either side will be poorer but still useful.  Call it a good 5x 2mm per transistor, 5x 8 mm for all 4.  This seemed a reasonable contact area and I was able to run 5.8W out key down for 20 minutes without any problem and the temperature somewhere in the midst of the transistor group  and ALSO the temperature in the bottom of the new copper heatsink was around 51c in each case with the top of my 30mm high sink fin still at around 40-45C.   If I assume the 'worst temperature in the midst of the transistors was 58C  I think that means junction temperatures for the BS170s of around 120C worst case which is allowable and sustainable.  

From this I concluded that it is possible to extract significant heat out of the curved tops of the transistors even though the transistors junction- to -case thermal resistance and / or the contact method used is not perfect for higher powers.

I would add that no one has suggested that running continuous 5.8W for 20 minutes is a very good idea, indeed my motivation is to be very sure a 3.5Watt beacon is bomb-proof in a heat wave.

I think the existing advice on limiting voltage and power in all the literature is good advice but if you wish to push the limits or raise the room temperature you can extract heat from above the packages