They are all very fast to make a comparison. I think your consideration would take into account whether blanking is needed.
For example, if sensing a low-side NMOS using Rds-On, the sense voltage starts high before NMOS turns on. This seems like a false OCP, so the signal must be blanked for some time (50-100nS) after the NMOS gate goes high. This limits your response time to a minimum of 50-100nS.
In contrast, sense resistor only senses the actual current through NMOS, so it's blanking time is much shorter or it may need no blanking at all - at each on time it's voltage starts at 0!. But sensing across a resistor contributes some IR loss.
You can insert a sense resistor in only a small portion of the mosfet to minimize IR losses, or use a sensefet that mirrors a portion of the output current. Both of these can become nonlinear quickly as the ratio of Switch to sense device becomes large so accuracy of current limit is degraded.