Heh... the caps is something I always roll my eyes at when somebody brings it up... myself included. If it is impacting the sound it definitely is not as intended, and absolutely is a sign that they are not functioning properly. I haven't looked at the S950 board in a while... so I don't remember if they have caps on the outputs or not... but that is frequently in place to help prevent voltage from going the wrong direction... ie... don't plug an audio output into another audio output... I usually see busted caps on the output chain just making the signal quieter... which is not something you want.
I just checked the
S900 manual on page 21, and it does say it only goes to a bandwidth of 16000hz. What I do not know is if this still equates to 48khz sample rate or is it the same 2.5 factor as the S950. The manual states that at 16000hz bandwidth you have 11,878ms of sample time... Assuming for a 48khz sample rate, this would mean the bandwidth works out to a 3x factor... but yes... this starts making the math funky. If I feel like it I might work on trying to figure it out later. To note, the numbers I posted above are with the Tone program/sample that load by default... and it's possible that the S900 (and probably S950) manual are ignoring the reserved memory for programs and the tone sample *shrug*.
I'd suspect that the S1000-S3000/XL range differences are likely due to anti-aliasing tech, and possibly rebalancing how the bits are applied. Plus don't forget that there is value in going from a 12 bit sampling engine to a 16 bit sampling engine... there will be more accurate (for various definitions of accurate) calculations, which could help explain why it sounds better... there might just be less noise in the system resulting from the 12 bit calculations.