I would probably look at current sense resistors at around 0.0005 ohms to 0.001 ohms instead or 0.01 ohm.
The reason is that if the resistor can heat up 60degC at full power, you can only expect about 0.3% accuracy and that would be a pity if you look at the cost. Buying very low offset opamps costs very little by comparison, and you can end up with a much cheaper current sense resistor and a much higher accuracy.
The transistor looks good, but devices with that kind of packaging are expensive compared to several devices in a TO247 package. It may be much easier to achieve your results with 42 100W modules, then your current 12 modules which will dissipate 280W on the mosfet and 70W on the current shunt. 280W means you have to keep those mosfets below 80 DegC and if the ambient can be up to 55 degC (as life gets hot near a 4200W heater), your heatsink has to be able to get rid of that 280 W per mosfet with an effective thermal resistance of less then 280/25 = 11 W/degC.
That is very ambitious.
At these sort of numbers, you can make your life much easier by just going for big numbers of much cheaper and more conservative lower power modules. A channel heatsink with a single fan and four mosfets can probably dissipate 200-300W. In a 19 inch rack width case, you make be able to fit 4 of these to get up to 1050W. Four of these cases can add up to the 4200W.
That would be a total of 64 65W mosfet modules. If each used a 0.001 ohm shunt resistor, a SMD 2W 100ppm shunt resistor costs about 50c, a IRPF250 is under $2, the opamp about $1 so you are looking at under $10 including PCB per module. The SMD shunt resistor would hardly warm up, so you will get about 10 times the precision of your current design.
Richard.