thermocouple multiplexing techniques
Here all depends from accuracy you want to achieve. RegUser_2 say well about antialiasing filter and different ground path from analogue and digital. As it was said the impedence is not a problem with thermocouples, because they are a quasi-ideal voltage generators (internal resistence close to zero).
Considering that you want measure temperature from thermocouples you don't need high speed, so you should use a delta-sigma ADC.
Delta-sigma ADCs are oversampled converters, so much less prone to aliasing problems due to noise "shaping" effects.
The market offers a variety of good Sigma-delda with 3 orders modulator, so you can avoid the use of prealiasing filters.
Some of such converters, like AD7730 (Analog Device) have a front-end amplifier with 10/20/40/80 mV Full scale, so very appropriated for direct thermocouples/RTDs measures. AD7730 has two multiplexed channels and a modality of Auto zero and Auto span of the whole measurement chain. Further, 100nA current generators used for continuity check of thermocouples are built-in supplied.
If you want a cheaper but still performant solution, you can select a LTC2418 from Linear Technology. You will have 8 differential inputs and by 24 bits high resolution and possibility to achieve a full range of full scale you can have a direct connection to your sensors.
To create the cold junctions compensator, you may use a single isothermal block and a single thin-film RTD included in middle position of block. You may connect this RTD to a free channel of ADC. Usually the compensating temperature range comes from 5 to 50°C, so in this range the RTD is linear, so you need not to linearize it, still achieving a cold junction error of 0.3°C (best case if you use class A RTD) or a worse 0.5°C. Once you have measured the cold junction temperature from RTD, you must convert this reading in the same equivalent voltage of thermocouple type J at the same temperature and finally you have to add equivalent voltage to FEM measured for that thermocouple. This is because all thermocouples linearization tables are calculated with cold junction to 0°C.
You will use these tables to linearize and scale the thermocouples readings.