In response to kendallcp: it works! I did it with tape back in the early '70s, and unknown to me at the time, Joe Dundovic of Nortronics (tape heads) was experimenting with it too. Tape goes "gracefully" into distortion when overdriven, and the only thing you have to watch out for with PWM is reaching the point when your duty cycle disappears and audio peaks slam the supply rails into the head. This can be obviated with some form of limiter ahead of the PWM circuit, or by comparing your audio waveform with a 'peaky' clock, rather than a triangle, to generate the PWM.
Another neat trick that works, sort-of, is to get rid of the AC bias altogether. It's possible to predistort the audio waveform in a manner opposite to the hysteresis-loop function of the recording process, although this requires fancy phase compensation as well as the nonlinear transfer. This gets rid of the "biased-tape noise," and really improves the high-frequency performance. The tradeoff is 'modulation noise,' which can be eliminated by applying a sort of 'dithering,' which puts the biased-tape noise right back. Such fun!
@betajet: Batman: I like to think it's because... our hearts are pure.
LOL It was so sappy -- I loved it as a kid. I remember whenever the Riddler created a riddle and they solved it and the solutuion was so wild and wacky ("... and what eats an unripe banana at three o'clock in the afternoon Robin...") but it all seemed to make perfect sense when you were a kid... you just thought they were really clever to sort out the clues :-)