Well, I made some tests:
I tried 3 frequencies at high duty-cycle, to reproduce the dwelling angle of the contact breaker, as the breakers remains closed most of the time (they only open when the rotor cam lobe hits the contact breaker arm).
Well, some time ago I had done those tests, as I was playing with a multi-spark ignition, but I lost my notes somewhere.
Well, like Speakerguy said, I could achieve 22 kHz. This can be proved with some engine math:
If I have a V8 with max RPM = 6000, I know that each turn on the flywheel has 4 power strokes.
Well, 6000 rpm = 100 rps = 100 Hz. But I have 4 power strokes so 100 x 4 = 400 Hz.
In a normal contact breaker engine, w/o multi-spark, the ignition coil would run at 400 Hz, but the engineer that projected the coil is a pessimist one, and he would project a coil that would operate well until 4 kHz.
Well, the tests results:
2.577 kHz, good spark, 15mm long (~19/32 "), D.C. =~ 97,86%
4.300 kHz, usable spark for a spark plug, 5mm long (~3/16 "), D.C. =~ 97,00%
9.689 kHz, minimun spark, 1,5mm long (~1/16 "), D.C. =~ 96,70%
The frequencies and duty cicles were measured with a oscilloscope.
Well, you could use the 4.300 kHz to modulate some sound, just for fun.