First off, I found out recently that the L2 does not measure "true peaks," so your signal may actually be going over that -10dB ceiling. Use the plain old L1 if you want to get a more respectable "brick wall" ceiling (though, to be honest, I'm not positive that it uses "true peaks" either). True peaking refers to issue of intersample peaks, it's a little bit voodoo, a little bit science, but beyond my abilities to explain further.

I usually set the output max at -10dB, and the threshold at -11 or -11.5

Second, if you're mixes are being ducked at broadcast, it's probably because the overall perceptual loudness level is to high. In other words, you're mixing too hot. The cheapest method you can use is the Phasescope plug-in that comes with Pro Tools. In the bottom left corner, you can select a setting labelled LEQA. Adjust the measurement time in the lower right hand corner and mix for around -27. This isn't by any means a precise method, but it will get you closer to actual broadcast spec.

Setting the ceiling at -10dB, and the threshold at 0, I find that my mixes are lowered 10dB overall. Why is my entire mix effected when the L2 threshold is set at 0?

Setting the L2 (or the L1 for that matter) like that is in fact equivalent to... lowering everything by 10 dB. By putting the threshold at 0, nothing gets limited.

For example, assuming that before inserting the L2 the max signal peaks at 0 dB, if the threshold is set to -10 dB on the L2, any signal going over -10 will be limited to -10.

After the signal is limited at -10, the ceiling is set to the target maximum peak value. If the ceiling is set at 0 dB, the signal will be raised by 10 dB. This will result in a perceived raise in loudness, but not a raise in maximum peak value since we started with a signal that already peaked at 0 dB.

In any case, you need to know the RMS and peak values that are accepted by a given broadcaster: some are more strict than others.

I'd say the Ceiling control is labelled wrong in the L series as it's not clear what it does, when in fact it's an output attenuator, and the Threshold control applies automatic gain compensation by as many dB.

I.e. for a threshold of -10dB, there will be 10dB of gain compensation. With the ceiling set at -10dB (attenuation), you're pretty much looking at just shaving the peaks above -10dB and the rest of your signal staying the same.

Threshold at 0, Ceiling at -10 = no limiting, no gain compensation, attenuation of -10dB.

For a broadcast setup: threshold at -10, ceiling at -10, but if you're feeling adventurous with your meters, both at -8... :)