There are a few different benefits from stacking buffers. One is higher output current, which is helpful for low-impedance headphones. This is what Tomo's referring to. I've stacked buffers to power low-Z unamplified PC speakers before, just for the hell of it...
But there's a few other factors to consider. One is the amp's output impedance. With one buffer, the output impedance of the amp is equal to the buffer's output impedance (10 ohms for the EL2001, 8 for the 2002). (Unless you add R8 or R9!) When you add a second buffer to each output channel, you divide the output impedance by 2. Add 2, divide by 3, add 3 divide by 4..., etc...
The lower the output impedance, the higher the damping factor of the amp, which increases the amp's control over the headphones, especially with the bass. I can personally attest that stacking a pair of EL2001s on the outputs of a META42 does make an audible difference with Senn HD-580s. (300 ohms) I tried 3 deep with these headphones, and that difference wasn't nearly as marked as the first level.
Another factor which I can't say much about is that stacking buffers reduces the load on each chip, which may help them to behave more linearly. Just a guess...
As for stacking the ground buffer, this is less important. However, it can help stabilize the rails. There is a procedure in the META42 documentation's tweaks section that tells you whether you need to stack this buffer or not.