Cgs should be pretty obvious: it does not allow the G-S voltage to rise rapidly, so it slows down the MOSFET. Remeber, the actual G-S voltage is what turns on the MOSFET, so if it changes slowly, the MOSFET turns on slowly.
The Cgs does basically the same thing, but since is is connected between the input and the output, its effect will be much greater. That is because of the so-called Miller effect. If an inverting amplifier has some impedance connected between the input and the output, then this impedance can be replaced by an equivalent one, between the input and ground, but of magnitude Z/(A+1), where Z is the impedance and A is the gain of the amplifier. Since the complex impedance of a capacitor is 1/sC, it follows that the Cgs ( it is the feedback impedance now) can be replaced with an impedance Z=1/(Cgs*(A+1)), that is, a cap A+1 times larger, between input (gate) and ground.
So, in essence,both Cgs and Cgd contribute to an equivalent capacitance from gate to ground that slowd down the G-S voltage rise and thus the MOSFET.
A cap across the gate resistor has the effect of speeding up the MOSFET. Without getting into detail, the cap provides a greater current to the gate during turn-on and turn-off. Basically, that cap transmits the edges of the signal more "directly" to the gate. That means more current available to charge the equivalent gate capacitance, which means the voltage across it rises faster, leading to faster switching of the MOSFET.
If you look at an oscilloscope probe, there is a trimmer you can adjust to "compensate" the probe. That trimmer does much the same job, it allows the probe to transmit the edges of the signal better, compensating for the input capacitance of the scope.