Sean's leaky boat analogy is not appropriate. He is all wet :)
If we must have an analogy, consider the task of steering your car around a curvy road. First you turn the steering wheel. Then your eyes notice that you are drifting into the opposite lane or into a tree. You apply a correction to the steering wheel. (Is that "negative"?)
This analogy also illustrates the limitation of a feedback control loop. If the car goes fast, and the road has sharp curves the driver may not be able to respond quickly enough to prevent a crash. In fact, if the required corrective inputs are too quick, the driver may become confused and apply inputs of the wrong polarity, and the car "spins out". For similar reasons. an audio amplifier with too much feedback is not only bad sounding...it becomes an oscillator.
Extending the analogy even further, some cars need less corrective feedback than others. A Ferrari corners better than Chevy. It also costs a lot more.
In a unity gain operational amplifier circuit (usually, but not always implemented by an integrated circuit device) the feedback is 100 percent. The FULL output signal is applied to the negative input of the differential amplifier. Since the desired signal is applied to the positive input, what drives the output is the DIFFERENCE between what you are getting for output and what you want. Important characteristics of the amplifier are slew rate and absurdly high gain. Linearity does not matter. (Low noise is nice).
In a normal linear amplifier some gain is desired, so 100 percent feedback is out of the question. Our audio amplifiers have bandwidth to at least 20KHz, and often to 100KHz. Circuitry like this does not delay the signal very much. Delay is bad for feedback because you need the feedback to be properly phased with respect to the signal. A given delay (microseconds) corresponds to more phase angle for high frequency signals. When the phase shift gets near 180 degrees your negative feedback becomes positive feedback, and you oscillate. The feedback signal is usually run through a filter to remove ultra-high (inaudible) frequencies where delay would induce oscillation. "Local" feedback (around just one stage of circuitry) is less affected by delay, and can be used to higher frequency than "global" feedback which is around the multistage amplifier circuitry.
When negative feedback is used, the forward amplification needs to "work harder". An analogy here might be a turboprop engine. The turbine of a 1000 HP engine needs to develop 8000 HP, because 7000HP is needed to run the compressor. (The exact numbers are my guess). This requirement, rather than the delay issue, probably accounts for sound quality degradation in very high feedback amplifiers.
An amplifier intended for use only with a subwoofer can have a very high amount of feedback, as long as high frequencies are removed from the feedback to avoid oscillation. Of course, in the case of subwoofers of the servo variety (eg: Velodyne) the driver cone is included inside the feedback loop, by developing a signal from a tiny accelerometer mounted on the cone. Here delay is a big time issue, which is why this technology has never been successful except in subwoofers.