For bypassing, it's a minimum capacitance rating, at which the noise on the logic board reduced to an acceptable level. The usual capacitance tolerance on modern electrolytic parts is +-20%. Further increased individual or total capacitance will slightly further decrease the medium frequency noise, but too much total capacitance (thousands of microfarads) will slow the power supply's corrections to any detected deviations from nominal voltage (increased deviation duration, and in extreme cases cause voltage instability: step load transient noise exceeding the 125 mV amplitude /5 mS duration spec). You have some latitude here, remember that the power system design has to tolerate the unspecified additional capacitance of any optional SE/30 PDS slot card bypassing. NuBus cards have a capacitance per card upper limit spec of 1513 uF on the +5V, 536 uF on +12V, and 698 uF on the -12V, for example, to keep the power supplies in an stable operating region of reliable regulation.
For coupling (logic board C3, C4) bigger capacitance is better (for bass response). For analog board coupling, bigger is usually better, unless the capacitor is part of a resonant circuit or used to linearize a waveform. On the analog board C9 can be bigger, while C15 may need to be low ESR but right valued for linearity.
Aluminum electrolytic capacitor manufacturers often offer a variant part type that is physically smaller than their standard parts for a given capacitance and voltage spec. These should be avoided for replacement purposes and most new designs because they usually will have higher internal resistance, reduced AC current capability, and shorter operating life.