As the head is smaller relative to the wavelength at lower frequencies than at higher, the shadowing is also smaller. Therefore there should be more blending of right and left below 1 kHz than at higher frequencies.

Crossfeed circuits are usually implemented in active headphone amplifiers which is an advantage as the frequency variation in the headphone impedance then won't play any role. However a passive circuit is much simpler, does not require power, and can be built into a smaller enclosure. Here is an adaptation of Siegfrid Linkwitz's passive circuit (little more than halfway down this page).

This circuit must be designed with the impedance of the earplugs in mind. I designed mine for the Koss Sparkplug which has 16 ohms. I also designed it for a low impedance output stage like in the iPhone 3G/3GS (3.1 ohms) or 4S (1.9 ohms). By the way I checked my iPhone 3GS for frequency variation across the band and it is remarkably constant, suggesting a capacitor-less audio output stage which is probably a digital class D design.

The passive crossfeed is shown below. Left_in and Right_out are connected with Common to a 3.5 mm minijack that plugs into the iPhone, while Left_out and Right_out are connected to a female minijack for plugging in the earplugs. The two Rin resistors, the inductor (Lx), the series resistor Rx, and the switch are the only ones required, as Rhead is shown only as a model for the impedance of the headphone.

The circuit has an inherent attenuation of Rhead/(Rhead+Rin) = 17.2 dB, so this means the volume control needs to be set higher than before, but in my opinion the iPhone still has enough output level. This value is also the high-frequency gain when the crossfeed circuit is on. At low frequencies the gain can be found from the voltage division between Rin and Rhead//(Rx + Rin//Rhead) where // means parallel connection. The values above give a gain of -19.3 dB, i.e. 2.1 dB lower than for the high frequencies. This is made up for by the leakage from the other channel which is at -30.5 dB, i.e. 11.2 dB below the direct channel. Note that Rx includes the inherent resistance in Lx also.The frequency response is shown below. The upper curve is from left input to left output and the lower curve is the response from the right input. With a mono signal as input, the two curves add up to a perfectly flat response.

The crossover frequency where the crossfeed starts losing importance is found where the impedance of Lx is the same as Rx, i.e. f=R/(2 pi L). In the circuit above this is 764 Hz.

The values above mimic the response of Linkwitz' circuit, only for a different value of the headphone impedance. In my case I wanted a bit more effect so I reduced Lx to 5.6 mH and set Rx to the internal resistance of Lx, 16 ohms, plus an external 10 ohms, ending up with 26 ohms. This gives 2.5 dB reduction at low frequencies and a cross coupling of -9.2 dB up to 7-800 Hz. The circuit is viewable in the CircuitLab editor.

Not all music needs this kind of filter. It is easiest to hear the effect on music which has been mixed after ideals of the 60's and early 70's when stereo was new (Edit: ... and when mixing consoles didn't have pan pots). This includes for instance early Beatles and Santana. But even some more recent tunes from U2 will benefit from crossfeed. And of course the application is not limited to the iPhone as it applies equally well to any mp3 or music player with a low impedance output.

Finally, if you want to understand more of why the shadowing effect varies with frequency, then take a look at William Hartmann's "How We Localize Sound" (alternatively here on the Wayback Machine). The top curve above mimics Fig. 2 in that article.