This also depends on incident power. A big limitation of CARS is sample damage. This is less a problem for single frequency CARS than it is for broadband, but no matter what you're always working on the edge of burning up your sample to get the best signal to noise.
Are you really going to do CARS for a student lab? That's impressive. Really any dense medium will work, especially organics with a high concentration of rings. If you aren't interested in imaging, then go straight for solvents. Toluene is great.