Well, you'll get as many references as opinions.
Safest: ask a manufacturer.
On the practical side of things, the voltage limit sucks because that necessarily prohibits high currents at low SoC -- yet you may need that for a UPS, or cranking an engine, etc. Note that internal resistance Ri goes up with discharge as well.
But maybe that kind of operation wears the battery extra, so it's best avoided.
As far as the chemistry goes, consider that the electrodes are made of spongy material, which greatly increases the surface area and thus capacity, but also greatly increases the distance ions need to travel through the electrolyte to reach any given point on that surface. Which means much more voltage drop through those pores (at high current density), and so they'll always lag behind the SoC of the outer/direct facing surfaces.
Perhaps the answer is both, because the outer surfaces could be utterly depleted (0% SoC) while the pores are still just waking up. Thus, wear proceeds inhomogeneously over the total electrode surface, and capacity drops as a result.
Capacity loss occurs (as far as I know), primarily because of formation of insulating PbSO4 crystals, and mechanically due to flakes/segments falling off or otherwise losing electrical contact (which might perhaps become broken by formation of such crystals, or by corrosion of the metal; note that material isn't plated back perfectly in place, with some deformation occurring over cycling, and this also limits battery life even under the most gentle use).
The porous structure also determines the dis/charge curve with respect to time; ionic diffusion is already a significant effect (giving a ~sqrt(f) impedance curve), but the distribution of pore sizes/lengths also gives an equivalent diffusion effect, and both effects stack to give quite long time constants -- hence the hours or even days cycle times, like for the float charge cycle having a long tail of current, dropping off slowly while held at fixed voltage.
As for references, let's see what the Google has to offer today...
You're welcome to play along, critically evaluating the veracity of these links.
Suggests AND condition, i.e. limit both voltage and SoC. Seems to be a user. At a glance, no references linked. Is this a primary source? Probably not.
Here's a shockingly familiar plot... no citations though. On the upside, it seems Richard was a clever guy: http://www.omagdigital.com/publication/?i=394240&article_id=2743496 unfortunately, we can't know what references he was pulling from.
12V Sealed lead acid battery level sensing
Links a datasheet -- primary manufacturer data -- though that doesn't include what cycle life the indicated cutoff point should give. Disregarding the lack of cycle life data, this gives a current-dependent cutoff, which roughly corresponds to an awareness of internal resistance: evidently about 0.6V/20A = 0.03Ω (taking the maximum current from each range).