Here is a more unconventional explanation in five steps:
We assemble a voltage divider...
Figuratively speaking, the voltage divider consists of two resistors "pulling" the midpoint (virtual ground) with equal force but in the opposite direction - R3 "pulls" it up and R4 "pulls" it down.
But the voltage divider is weak...
The problem is that the resistors "pull weakly" (because they have relatively high resistance) and when another device starts to "pull" on that point, it gives in and starts to "move".
The straightforward solution
is to decrease both R3 and R4 resistances... but then the voltage divider will consume a lot of current.
The ingenious solution
can be found if we begin looking for a more clever solution. Thus we arrive at the idea that it is not necessary for both resistors to be of low resistance; only the "opposite" resistor should have low resistance. For example, if some device is "pulling" the virtual ground up, R4 should have low resistance to "pull" hard it down and v.v., if the device is "pulling" the virtual ground down, R3 should have low resistance to "pull" hard it up.
That is how we get to the idea that the resistors should be dynamic.
We implement the idea
This idea is implemented by the op-amp output stage which is inherently such a "dynamic voltage divider" consisting of two transistors (the so-called "push-pull" or complementary stage). It can be thought as a "low-resistance copy" of the R3-R4 voltage divider.