fireonshakedwnstreet and SMsound, thanks for the replies.
I've been quite puzzled about where the other interpretation of "headroom" came from, where people think that it's a more or less wide zone below 0 dBFS that should be kept clear altogether. I think that it might come from the very dishonest polemics against digital audio that many audiophile magazines waged in the 1980s; they used to debunk (or so they pretended) 16-bit PCM's claim to have greater dynamic range than analog recording, in part by subtracting 10, 12 or more dB from the top, claiming that this was the "headroom region" and shouldn't be counted. (Of course, though, they counted every last dB of headroom for analog recording, despite the sharply increased distortion that occurs in that region.)
SMsound, there's no one norm for limiter behavior, and a circuit that starts to restrain the recorded levels as low as -7 dB is a bit of a "soft knee" in my book, between being a limiter and a compressor. I'm more used to "harder" limiters that kick in just below the limit of the medium, and that hold the line there. In any case I agree that anything more than the smallest amount of limiting will change the character of sound, and I don't want any non-emergency limiting in any live recordings that I make. It's better to under-record by 6 - 8 dB than to over-record by 3 or 4, especially if the system doesn't overload gracefully (some do and some don't). Also, with stereo recording it's very important that any limiters be "ganged" between channels. The limiting action, if it occurs, shouldn't cause the stereo image to shift left or right; that's more distracting than momentary "clean" overloads.
--Classical and opera are what I mainly record as well, but even with a 24-bit recorder and quiet preamps, I'm pretty sure that -18 dBFS for one's "all-evening" peaks is lower than optimal, especially if your recordings are ever played back in analog.
But circumstances do vary. It's impossible to calculate accurately with so many unknowns. Preamp noise depends greatly on gain settings, for example. The published noise specs for nearly all preamps are obtained at their maximum gain settings (50 - 60 dB or even greater), where the equivalent input noise is nearly always at its lowest by a significant amount. In concert recording I usually run preamps at ca. 30 - 35 dB gain. Around 20 years ago I measured the noise of all my preamps at ca. 30 dB gain; their relative rankings came out very, very differently from full gain. One of the quietest preamps at 30 dB gain, for example, was the noisiest one at full gain by a considerable margin.
And the noise spectra of condenser microphones vs. preamps are typically quite different, as is the prevalence of "shot" (impulse) noise vs. smooth, continuous-sounding noise. All in all the total, effective, potentially audible noise that you'd get from any given combination of microphone + preamp ... as I said, it's tricky, and back-of-the-envelope arithmetic isn't up to the task. So I stick with the accustomed practice of maximizing the signal levels at each step along the way while taking care to avoid overload. Nothing fancy, just basic principles.
-18 dB, incidentally, seems to come from Sony's old recommended practice for the PCM-1600, -1610 and -1630, which was to set 0 VU = -18 dBFS on steady tone. (I worked quite intensively with those processors in various East Coast recording and LP mastering studios back in the day.) But that recommendation was based on realizing that VU meters are "syllabic" rather than peak-responding, and that people quite properly let them go up to +3 VU for momentary signal peaks. The result with actual program material is that signal peaks on the digital side would commonly reach about -6 to -3 dBFS. Back then, dither wasn't widely understood and with many recorders (including the professional Sony processors that everyone used for CD mastering, in their default settings), distortion and "granular noise" increased quite audibly at lower signal levels. There was never any intention of leaving a wide zone unused near the top of the range!
All told, to me it seems very likely that there would be some s/n benefit if you raised your levels at least somewhat, especially if you consider the noise of your recorder's (or any media player's) analog output circuitry.
--best regards