Drive levels depend on what the mixer actually is, and how much signal it needs.
...Duh?
Well...
If it's a DBM type, then it needs enough LO power to forward-bias the diodes. To behave as a toggle switch, this must be at least a few times the maximum RF input (I believe +10dB is typically suggested), and needs to be from a low enough impedance to drive the diode capacitance at the desired frequency.
Note that the source impedance is used as current limiting; all the LO is doing, is driving current through some diodes. So as those diodes conduct, they will reflect a lot of power. So, expect a poor VSWR. Which is confirmed with the datasheet. Also expect more drive power (and perhaps a lower LO impedance) for higher operating frequencies. Plan accordingly!
If it's a transistor type (e.g., CA3028 single balanced; MC1496 fully balanced; anything JFET or MOS, from single or fully balanced bridges to analog switches), then the goal is again to deliver enough voltage to achieve good mixing, while using a low enough source impedance to drive the capacitances to that frequency. But, there is no dependency on RF input power (or at least a lot less). The drive power can be much lower in these cases, which you can imagine as obtaining amplification from the transistors, or at least getting light loading from FET gates. BJTs only need ~0.1Vpk to nearly fully switch on/off, with junction capacitance similar to diodes'. FETs are mainly capacitive, so the resistance can be quite high, though if the capacitive reactance has to be canceled with inductance, the LO frequency range might not be wide (BW limited by narrowband tuning).
Note that "system impedance" is simply what's handy, and what they measured it at. YMMV, for better or worse, if you wish to try other impedances. If you can't take the time, follow the datasheet; if you can, there are probably gains to be had (power efficiency, noise floor, dynamic range).
Tim