Your questions are, as usual, 'spot on' and in tune with my way of thinking.
While there is evidence (some of it highlighted by Pierre and Bill) of degraded performance at high levels for normal hearing listeners, I would argue that the degradation is comparatively less for normal-hearing listeners than for hearing-impaired listeners, even though as you said normal-hearing and hearing-impaired listeners share common cochlear mechanical properties at high levels. Also, a paper just published (Gregan et al. 2013 JASA) suggests little correlation between residual compression and masking release. Therefore, I think that reduced compression and broader filters can explain only a (small) part of the difficulty experienced by hearing-impaired at understanding speech in noise. The question is what explains the other (larger) part. I think the answer is 'deafferentation'.
We know that age is a separate contributor to degraded performance in noise that adds on to hearing impairment. This is beautifully shown by Table I of Peters et al. (1998, JASA 103:577-587). What I find most interesting, however, is that age leads to auditory nerve deafferentation (Makary et al. 2011, JARO) and deafferentation also occurs after noise exposure (see Kujawa and Liberman 2009) even in clinically-normal individuals. Therefore, it is very likely that deafferentation occurs even more frequently and significantly for hearing-impaired individuals. We have recently shown (Lopez-Poveda and Barrios 2013 Front. Neurosci.) that degradation of the acoustic signal inspired by deafferentation can impair speech intelligibility in noise but much less so in quiet without altering audiometric thresholds. Altogether, this makes me think that deafferentation is a common, underlying cause of degraded peformance in noise for both hearing-impaired listeners and (audiometrically-normal) aged listeners. What is more, I would dare saying that deafferentation degrades performance in noise more than does reduced frequency selectivity (broader filters) and/or reduced compression. The mechanism of how deafferentation would degrade performance in noise is described in our paper (Lopez-Poveda and Barrios 2013 Front. Neurosci.).
That's my current thinking, anyhow, and here are some relevant references: