Other options:
1) Spread the heat over larger area by soldering (or other ways of non-isolating mounting with very low thermal resistance) the device to a copper heatspreader (like a 50x50x5mm copper block), this directly reduces the thermal requirements for the isolating material by increasing contact area, so very typical 2-3 W/mK 0.5mm thick or so silpads could be used.
2) Throw more silicon at the problem. If the IGBT is of easily parallelable positive Vcesat thermal coeff type, this would be obvious. Yes, it increases the silicon cost, but by cutting the per device dissipation, you might even be able to reduce heatsink size (and cost!) and run the heatsink hotter while maintaining acceptable Tj. Not saying this is necessarily the case here, but I do see the trend that today, silicon is often cheaper than aluminium, so to speak!
My go-to strategy nowadays seems to be, optimize for highest efficiency, use more silicon, and/or better silicon; power not dissipated at all is easiest to heatsink! Energy savings are a bonus.
I'm mentioning the other options because having to think about the isolation layer this hard may be a sign that you are running pretty close to the limits.