True, the surround channel delay is supposed to be implemented by playback hardware only. And it is only supposed to happen during upmixing, not when playing 5.1 mixes.
In this case, the ‘hardware’ is the Dolby Media Decoder application that created the upmix. So for this scenario, it takes the place of the upmixer in the receiver, and the receiver only sees a 5.1 mix. It has no way of knowing that it originally came from a 2-channel source, and so will not apply any additional surround delay beyond what it is already doing to time-align the speakers. Thus when playing it back in 5.1 format, it will sound the same as if the receiver had done the upmix, aside from the addition of the LFE channel. (The LFE channel is, of course, the only reason for even bothering with making it 5.1 in the first place; if not for this, I would have distributed my edits in stereo and just let people upmix them in their receivers like they would anything else.)
So really the only part that is up to interpretation is how much surround delay should be applied in the Dolby software, since it gives full control over parameters which are normally hidden in most consumer equipment. I used longer delays in Empire and Jedi specifically to minimize the comb-filtering issue when the 5.1 is downmixed, since I knew I could not prevent people from doing this no matter how much I urged them not to. The only way to eliminate it completely would be to apply no surround delay at all, which is not an option in Prologic II movie mode or in the original Prologic. Only Prologic II music mode offers this option, but the lower channel separation in this mode is not ideal for film content. I could manually compensate for the delay in Pro Tools after recording the upmix, but then we’re back to the problem of crosstalk influencing the listener into thinking sounds that are front-panned are coming from the back of the room. Dolby specifically designed their movie upmixers to take advantage of the Haas effect, and since my 5.1 is an unusual case, I realized I had to follow their principles as well as I could while taking the differences into account.
Anyway, like I said, I believe I’ll be able to further reduce this in a subsequent version to the point that it will no longer be an issue, taking advantage of the audio engineering experience I’ve gained since then.