If you don't bond the neutral, then you are quite correct: a ground fault _won't_ cause the OCPD to trip.
On the other hand, if the secondary of the transformer is not ground referenced, then a _single_ ground fault won't have much if any current flow; this is what is known as an 'ungrounded system', and some systems are intentionally designed this way. Dig though the NEC and section on grounding and you will see that some systems are simply _not_ required to be grounded. The idea is that the system will continue to function with a single ground fault. Generally the way that this works is that you have circuits set up to detect ground faults, and in the event of a ground fault you work to isolate the faulted circuit, then actually shut the system down at a convenient time.
There are many problems with this approach. In particular, you always have _some_ coupling to ground, even if it is only capacitive coupling. In the event of an intermittent ground fault, you can actually develop phase to ground voltages that are several times the normal phase to phase voltage; in your example, it doesn't really matter if your 480V HVAC circuits are 277V, 277V,277V or 0V,480V,480V, but it very much does matter to the insulation system if you ever manage to hit 1600V,2000V,2000V
The 'modern' approach to this sort of system is the 'high impedance grounded' system, where the neutral is bonded to ground via a resistor. The resistor is sized to permit a significant but not particularly large current to flow in the event of a ground fault; enough to solidly tie the voltages to a ground reference, and enough to detect the ground fault, but not enough to cause significant damage. The system that I work on this past spring used a 50 ohm resistance to ground on a 480V/277 system, giving about 5A of ground fault current.