5 years later, and I find this question on Google. So, I'll add my comments for the next guy looking for info.
There's lots of documentation describing the selection of R2 and R3, and the capacitors, but little documentation on the effect of R1 and R4.
I'm playing with my scope and a multivibrator circuit (without steering diodes). I've noticed that my waveform is much more sawtooth shaped than square when I'm using 1k/10nF RC networks and 270R collector resistors (at about 70kHz). However, increasing the RC network to 51k/100pF, and keeping my 270R collector resistors really sharpens up the waveform (at about 150kHz).
Here's the 51k:270R

And with 1k/270R

About the best analysis I can come up with is this (and I might be wrong):
These transistors are operating in common-emitter mode. Therefore, gain is proportional to the ratio of the resistors R1:R2 and R4:R3. My transistors (BC337-40) should be capable of a gain of 250, but using 1k:270R sets the max gain permitted to about 4. A higher gain would drive the transistors to the rail (saturation) much faster than a lower gain. So using a 51k/270R configuration sets the gain to ~200 and reduces the rise time, sharpening up the wave.
Think about it this way - as the capacitor comes up to voltage, that increase is multiplied by the gain of the circuit, until it hits your V+ rail (saturation). If the gain is low then the increase is also low, and there's a broader zone where the (capacitor voltage * gain) is still less than V+.
So if your goal is sharp square waves, you should use high beta transistors and the highest ratio of R1/4:R2/3 that you can manage. Steering diodes may or may not be necessary depending on just HOW square you need your wave to be, the frequency, and the particulars of your transistors.