No, I'm distinguishing between matrix decoding (where a centre channel is extracted from other channels) vs downmixing (where the centre channel is combined with other channels).

Then we have a difference in terminology. I consider downmixing the (usually linear) mixing of upstream data to be sent downstream, as in a pipeline process. The number of channels need not be reduced. Thus, a three channel to three channel downmix is possible.

What you say about extraction is true, but, because PLII extraction permits control over the distribution over the front three channels, I separated into a fixed extraction (which is likely a non-linear transformation) and a linear mixing transformation. It is the latter that I would like to see provided in general.

You may have a problem with your system.

Since I only have a 2.1 setup at the moment, I don't have a system to have a problem with. However, your observation is interesting, particularly with PLII extraction (I would expect a discrete mix to fare well).

Are you suggesting that a PLII extraction to two front speakers (no centre) is no different than a PLII extraction to three front speakers, including a fully extracted centre (i.e. what would be a "normal" extraction setting)? If so, how far apart are your two front speakers? My 520i sit 12 feet apart, and, on stereo test material, can image quite precisely between them, and to some small degree, outside them, at the sweet spot (not that this is surprising).