They are both cheap. There is a piece of PCB balanced on top of the K-type thermocouple's probe; there is also a red laser light from the infrared thermometer right in the middle of the same PCB where the probe's head can be seen. The difference between the readouts is sometimes larger than 100 degrees C. Which one should I trust more (for the purpose of PCB oven soldering) ?
Thermocouples are more likely to give an accurate reading (of the junction temperature). The relationship between the junction temperature and what you are trying to measure is another matter, but it's unlikely to be really disparate if you are careful (for example, extending the thermocouple wire inside the heated zone and coiling it against the PCB to act as a heat sink to keep too much heat from flowing away from the junction). A cheap thermocouple is likely almost as accurate as an expensive thermocouple (there are a few grades of wire, but even lower grades are pretty good). The construction (thin wire or ribbon is better for not affecting the temperature being measured) is more important. The linearity and accuracy of the readout device is sometimes pretty suspect if they're cheap, but they're usually good to 1% or better.
IR sensors give a reading that's highly dependent on the emissivity of the target. If you aim it at a specular surface you may get more of a reading of the walls of the enclosure or whatever else is reflected in the surface than the surface itself, which can be grossly in error. Certainly if there are open glowing-red elements within reflected view of the IR thermometer you're likely to get large errors on the high side.
Cheap IR thermometers assume a fixed emissivity of 0.9x, so something dark (in IR wavelengths) and matte will give you a more accurate reading. They also have variation depending on the sensor temperature so avoiding gradients will help. I would not trust a cheap IR thermometer very far, and even expensive ones have similar limitations to some degree.