My perception has been that Ralph is often not here on weekends, so I’ll give your question a shot while we await his more experience-based response.
The rough approximation you mentioned earlier to the effect that the 350 to 2000 Hz range within which your 100 db/W driver will be operated means that roughly around 25% of the system’s total power requirement would typically be supplied to that driver seems to me to be a reasonable ballpark assumption, for many and perhaps most recordings. If we assume that the amp is reasonably well matched to the driver’s impedance, and can put out a maximum of about 1 watt, the resultant 100 db maximum SPL at 1 meter would correspond to about 90.5 db at a listening distance of say 3 meters, neglecting room effects. Two speakers can be expected to raise that by 3 db (and potentially by 6 db, depending on listening position, but let’s be conservative and assume 3 db). So that brings us to 93.5 db. As a very rough guess let’s assume that room reflections would raise that by another 3 db, to around 96.5 db. Given the 25% assumption, outputs from the other drivers would raise the total maximum SPL to around 102.5 db.
To avoid operating the amp above say 25% of its power capability (Ralph mentioned numbers equivalent to 20% but using 25% eliminates the need for me to pull out my calculator for the subsequent calculation), 6 db would be subtracted from that amount, resulting in a maximum SPL at the listening position, with the amp operating in its presumed comfort zone, of about 96.5 db.
That will be good enough for the great majority of recordings for the great majority of listeners, IMO. However recordings having particularly wide dynamic range, such as well engineered minimally compressed classical symphonic recordings on labels such as Telarc, Sheffield Labs, Reference Recordings, etc., could very possibly put the amp out of its comfort zone, IMO, and perhaps even occasionally drive it into clipping, on very brief high volume dynamic peaks. Even though those recordings might be played back at average levels of perhaps 75 db or so.
I have many such recordings, that I listen to fairly regularly, so for me a system must be able to cleanly generate SPLs of 105 db at the listening position to be acceptable. For most recordings, genres, and listeners, however, I suspect that a maximum capability of 96.5 db would be fine. Especially given that the amp could provide 6 db or so of additional capability if occasionally necessary, albeit with some increase in distortion.
For an example of a recording having exceptionally wide dynamic range (although in this case the highest volume levels are reached on notes having much of their energy in the deep bass region), listen to the last five minutes or so of the ca. 1979 Telarc recording of Stravinsky’s Firebird Suite, Robert Shaw conducting the Atlanta Symphony Orchestra:
https://www.youtube.com/watch?v=Y5Or4x6RXyU. (Feel free to ignore the images).
Note how the volume of the music descends to barely a whisper shortly before the start of the finale at 17:25, and then rises to a concluding note that with the recording played in my system at average levels of perhaps 75 db or so closely approaches 105 db at my listening position. The 30 db difference means that around 1000 times as much power is required to produce that concluding note as is required to produce the average level of the recording. And based on SPL measurements I have taken, and on waveforms I have had occasion to look at on a computer (using an audio editing program) for other recordings having comparable dynamic ranges, I feel fairly certain that the difference in volume between that note and the softest notes just before the finale is upward of 50 db. Meaning that more than 100,000 times as much power is required for that note as for the softest notes.
In contrast, my understanding is that many and perhaps most pop and rock recordings are dynamically compressed to less than 10 db, meaning less than a 10x difference in power between the loudest and softest notes.
So given also my belief that perceived "loudness" tends to be mainly a function of average volume rather than peak volume, and given that the system you are building should easily be able to handle the average listening level of 85 to 90 db that you mentioned, it seems to me that the dynamic range of the music that is listened to is what should be considered as having the potential to drive the amp out of its comfort zone, if not to its limits.
One further comment: Consistent with one of the points Ralph stated earlier in the thread, in the absence of detailed technical information for the specific SPL meter that one may use I would not assume that the meter and its microphone necessarily have either the speed or the frequency range to fully capture the maximum volume of a brief musical transient.
Continued best of luck with your project. Best regards,
-- Al