It sounds like you should float a whole circuit at the 3 kV level. This would probably be something with a small sense resistor and active circuit to amplify the result. That would be presented to the built-in A/D of a microcontroller, which would regularly report the value serially over a single opto-coupler. If you really really need it to interface via IIC to the rest of the system, that can be done with a micro on the ground-referenced side that receives the serial stream from the opto and presents itself as a IIC slave.
Keep in mind we're talking about cheap micros here. The opamp will likely cost more than each micro.
If you don't already have a small power supply referenced to the 3 kV high side, then you'll have to get power to the circuit there somehow. I know this may sound inelegant at first, but how about batteries? If this is occasionally run in a research setting, then changing batteries every few months or a year will be less time and money spent than designing and implementing a 3 kV isolated supply, even if this device lasts for a few decades.
If you can't use batteries, then a transformer is probably the best choice. You will likely have to wind it yourself. Fortunately, the high side circuit should only need a few 100 mW, so it doesn't have to be all that efficient. Maybe a suitable toroid potted in high voltage goop. You can run the primary open loop with a fixed amplitude and frequency square wave. This gets full-wave rectified, then run thru a small linear regulator to clean it up and make a nice steady voltage. Again, efficiency shouldn't be a big issue, so you can go for simplicity and robustness.