Multichannel sounds can be positioned in space using the 3D positioning tab just as well as mono sources, without downmixing to mono.
In order to compute each source's channel's contribution to each speaker, "virtual emitters" are laid on a semicircle around the listener. The extent of that semicircle depends on the spread (100% is the whole circle, 0% is a point in front of the listener). This semicircle then is split in as much regions of equal size as there are channels, and the source's original channels each contribute to their assigned region. For example, with a stereo source, all virtual emitters on the left of the circle are constituted of the source's left channel, and all those on the right are constituted of the source's right channel. Once this is done, the actual contribution of each virtual emitter is computed for each speaker. Thus, the multichannel image is somewhat preserved even with 3D positioning.
Note that in the degenerated case where the spread is 0, all virtual emitters are located in a single point in front of the listener, and this is therefore similar to downmixing all channels to mono before positioning it in the 3D space.
