Are you saying that following this curve's natural response profile yields worse results? I said flatten earlier because it looked flat to me, but maybe my eyes are deceiving me.
There are lots of facets to response curves. The one Sigberg posted above is probably not a measurement with unlimited duration. I guess there's some frequency dependant gating going on, meaning he has cut the duration for each frequency band to something he found useful. That means there may be more energy in the room that is not included in the measurements. If this was included, it would most likely skew the result towards more lift in the low mid an bass.
I am not sure what info is available on how ARC does their measurements. There are countless ways of doing this, so unless we have that information we can not determine what they are actually doing.
A correction system has to make some choices. What should be corrected and what should be left untouched? How much do we correct for in each case? The curve may just as well simply show what ARC has chosen to correct for.
The topic of room correction is a big one and it is not what this thread is about. However, there are some possibilities built in to the Sigberg speakers, so a part of it will be on topic.
Firstly, any in room measurement will have some kind of mixed phase response. Once we start correcting for that, we introduce new errors in the time domain. That is regardless of whether FIR or IIR is used, but the errors can be a bit different. The trick is to just correct enough to dampen the bad stuff a bit. We can look at it in four segments:
1: Direct sound is mostly just the anechoic response of the speaker. Most of what we see here is minimum phase that can be safely corrected for. This is what we measure when the time window is kept so short we are not affected by reflections at all.
2: Reflections comes in when we extended the time window a bit. This is not something we should try to equalize as it is almost exclusively excess phase (so really bad time domain errors will be introduced) and it is not globally valid. That last bit means of we move the head a bit the correction no longer fits the sound it is trying to correct.
3: If we extended the time window further, we include decay and room modes. We should be careful when trying to correct for this. A significant peak that can be found in different locations can be turned down a bit, but not 100%. That would completely kill transients.
4: Below a certain threshold, commonly referred to as the Schröder frequency, you can more safely correct for modes. This is typically below 200-400Hz, depending on the room size. Corrections in this range does not degrade transients as much, and the time domain errors will be much smaller since the speakers and room behave more like a unit here, and we do not identify the speakers response separately from the room's response (like we do at higher frequencies).