There seems to be some cross-purposes discussion going on here - it's not real disagreement, just ... just varying priorities?
Smashwords may have a great automation system that works most of the time - or perhaps one that people have learned to work within. None of that denies the fact that the use of one-size-fits-all automation is always going to have some "edge" cases. One of the problems with such systems is that, as an outsider, you don't know if you're an edge case until your book turns out wrong (hence that separate thread looking to testdrive Smashwords).
One way to avoid many of these edge cases - which will probably exist regardless of how clever they make their automation - is to support something like epub directly, even if that means that there are some limitations imposed on what they do with it (what they convert to etc.). It seems like an obvious, and fairly easy, thing to do, so I'm guessing there's some reputation covering stuff involved in why they don't: by forcing everything through the same grinder they limit how the system can be abused, and so limit how they can be blamed for poor products (blame from either end). But they might find that such a reluctance to look ahead will backfire on them in the long term.