There are several schools of thought about this - Dennis's approach seems to be the most thorough, but others (including those with credentials) claim that less is sufficient.
Best I can tell from reading about this, Dennis and Shawn would set up the QSC with both a delay and also an all-pass filter. (This is entirely inferred on my part, obviously I'm not speaking for them, or from experience.) The QSC unit, different from most others, has a fixed latency, which makes calibrating multiple channels very easy, compared to other less costly units where the processing delay adds up, making final alignment more challenging to determine. They may also use some other processing, but I don't know what it would be.
BIG posted just the other day that Dr. Floyd Toole - noted and sometimes controversial researcher - spoke at CEDIA and claimed that a 10ms delay was adequate.
Other forum members claim that they just split the signal and carry on, very happy with the results. Others split the signal and then adjust mounting position and/or level so that precedence keeps the image where they want it, again very happy.
If you want to emulate the performace of the QSC for less, there are other choices, like some Xilica procesors which get good reviews from professionals, but lack fixed latency. Then there are less costly choices - in particular miniDSP. I think
an all pass filter
, which changes phase based on frequency, requires FIR functionality. Only certain products from miniDSP are capable of FIR filters, so shop carefully, if that is your goal.
Sorry I don't have links for all these claims, but I bet someone will come along and either back me up and provide some links, or clarify where I'm wrong.