I'm not familiar with the online calculators you referred to, as I just calculate the numbers myself, when necessary. But while the 100 db figure you cited is theoretically possible, I'd be a little more conservative and call it 97 db, as calculated below. 100 db would be realized under the conditions you described, for non-planar (box type/dynamic) speakers, if:
(a)The signal is such that 25W is simultaneously applied to both channels.
(b)The outputs of both speakers sum together in-phase at the particular listening position.
(c)The speakers can handle that power level without significant thermal compression.
Others may see it a little differently, but what I do is:
(1)Convert the power rating to db relative to 1 watt, on the basis of 10 x logarithm (P1/P2), which in this case would be 10log(25/1) = 14 db.
(2)Add that to the SPL produced by the speaker at 1 meter in response to a 1 watt input. If the speaker's impedance is 8 ohms, the SPL produced in response to 1 watt will be the same as the response to an input of 2.83 volts (which is a common basis for sensitivity figures); if the speaker's impedance is 4 ohms, the 1 watt response will be 3 db less than the 2.83 volt sensitivity.
In this case, 90 + 14 = 104 db.
(3)Add 3 db to reflect the presence of two speakers, rather than the 6 db which is apparently presumed by the calculator you used (and which will only be realized under the optimistic assumptions I described above).
104 + 3 = 107 db.
(4)Subtract the attenuation that occurs as a result of the listening distance being greater than 1 meter, based on 20 x logarithm (Listening Distance/1 meter). That corresponds to 6 db of attenuation per doubling of distance. In this case, 20log(3/1) = 10 db. (That amount of attenuation would be considerably smaller if the speakers were planar).
107 - 10 = 97 db, as I indicated above.
Some suggest that amplifiers should have x dB of headroom to cope with transients. 10 dB is often cited. Any comment on that?
Not sure what the 10 db is relative to. If it is relative to the average listening level of a particular listener, then your concluding statement is most likely applicable -- it will be adequate for some musical genres but not for others.
While I agree the dynamic range of a recording will set the pace on what multiple of power is needed, few of us will know what that recorded dynamic range is?
What I've done to get a feel for all of this is to purchase a Radio Shack SPL meter (a digital one in this case), set it to C-weighting and to the fastest response time setting, and measure peak SPL's on a number of recordings that I can tell by listening are among those having the widest dynamic range in my collection. In my case those are classical symphonic recordings on labels such as Telarc, Sheffield, and Reference Recordings.
I've found the highest peaks on those recordings to fall in the 100 to 105 db area at my listening position. As calculated per the methodology I described above, my system can produce 108 db at the listening position. So I have a little bit of margin relative to the worst case peaks (at least 3 db, which is a factor of 2 in power), and lack of knowledge of the recorded dynamic range therefore becomes irrelevant.