Quite apart from the SOA issues noted, it's likely that a pair sharing a nominal load current are ach working in an area of greater current-gain (hfe on the datasheet - there's often a graph on a good datasheet) - which means the VAS/driver stage 'sees' less of a load, even though it is now driving two pairs not one. And in addition, the effective base drive input capacitance (which can be really quite large with BJT outputs when a pair is pushed, c.10nF in some circs) is so reduced.
Together these are a good thing - less open-loop distortion, a hair more forward gain, and so - being a feedback-defined amp - a more accurate output is possible.
Edit to add: 'if some is good, is more better..?' Not really for amps like this: adding a third pair will be down in the mud, because the big step is going from one to two well-driven output pairs. If your application needs more output current than that - you probably need to re-think the preceeding stage to take advantage of it / or simply put - a 'bigger' amp in the first place.