Thanks for the confirmation guys. I'll change the LAV Audio Decoder settings to pass the full bit depth to Reclock, and reduce as needed there. (The 48/24 problem has me stumped. I'm able to play 96/24 files in foobar using WASAPI, so I'm surprised that trying to output 48/24 from Reclock generates an error on playback. I'll have to try to track down a 48/24 audio file to play in foobar to see if the problem lies in my MPC-HC setup and not in my hardware).
I did a little more reading about the "Clipping Protection" and "Normalize Matrix" settings, and it appears that both of those settings attempt to compensate for the increase in levels that naturally results from folding 5 or more channels of audio into two channels. Turning on Normalize Matrix appears to, well, normalize the levels of the resulting stereo downmix to preserve the original dynamic range of the multichannel source material.
Clipping Protection, on the other hand, seems to take a "wait and see" approach: the stereo downmix is output without normalization, meaning that the output levels will be "louder" now that material from five channels was combined into two channels. That increase in levels runs the risk of clipping, and so if Clipping Protection detects an impending clip in the audio during playback, it dynamically reduces the total volume of the audio output. I think the effect would be that if a loud explosion or similar effect occurred that would cause clipping, the viewer would momentarily hear the overall volume level drop during that passage to prevent the clip (which would be a bit of a weird effect I'd think).
I was scared of the "Normalize Matrix" setting from the two-channel audio world, where normalization usually means compression of dynamic range or other lossy manipulations, but it seems like that's the best setting here. nevcairiel said Clipping Protection was provided as an alternative to Normalize Matrix only because:
some people complained that the resulting audio [after normalization] was not loud enough for their low-powered speakers/headphones (because the volume of all channels is reduced to remove the theoretical possibility of clipping)
The only disappointing thing is that LAV Audio Decoder doesn't seem to pick up on downmix metadata that may be present on a Dolby Digital AC3 stream (or others, I'm sure, but the references I saw were to DD). I learned that audio engineers creating a DD soundtrack have the ability to embed the level at which each channel in the multichannel soundtrack should be mixed when downmixing to stereo. LAV's default settings correspond to what seem to be the "default" Dolby specifications for downmixing when the audio engineer doesn't specify coefficients, but LAV doesn't appear to have the ability to pick up on those mixing coefficients if they are present in the AC3 metadata.
http://forum.doom9.org/showthread.php?p=1555028#post1555028Edited by marvin4653 - 6/27/13 at 6:28pm
Note that some audio formats allow for metadata containing custom downmixing coefficients. I don't think LAV Audio uses them though, which is unfortunate.