Besides e.g. adding a VPULSE of fractional mV in series with one capacitor, you might set initial condition on one of the capacitors to a non-default value, so that it starts up unbalanced, rather than having the simulation engine calculate the DC operating point all too perfectly.
The hFE used in the 2N3904 is surprisingly high (over 300), but don't let that fool you, real ones are more like 200 or 100, and even less if you want saturated behavior (which you do for this circuit). A ratio of Rb / Rc <= 50 is quite reasonable here.
One final catch: SPICE doesn't model BJT breakdown (IIRC, LTSpice doesn't either). Note that, when a transistor switches on, it pulls down the capacitor on its collector, pulling down the base voltage of the opposite transistor by the same amount (nearly). For a 12V supply as shown, you're asking for pulses of about -11.3V on the base of those transistors -- but they're only good for about 7V!
To model B-E breakdown, connect two 6V zener diodes in series (use the nearest SPICE model in the library, it'll be fine), anode to anode, and connect them from base to emitter. Do this for each transistor. You will see a higher-than-expected operating frequency, a slower collector falling edge (because the capacitor has to be charged up by about 5V), and a flat-bottomed base voltage.
Best solution is to run this circuit at or below 5V, and if you need a 12V output, add an inverter as level shifter.
Tim