Fortunately, I was dropped on my head three times, so I'm in
reality C. I run epubcheck, learn that attribute "name" is not supported in XHTML and should be changed to "id", and make my ids unique, and learn it for my next books. After my book passes epubcheck I'm fairly confident my book is valid ePUB, and now start checking for software glitches. I have a quick look in a web browser, and in calibre, then upload it to my Cybook and read the book. After some days or weeks of swearing over ADE's glitches, stupid Cybook's margins, incomplete Unicode font support, etc. (this is the time it takes me to "proofread" the book), I have a book that I find satisfying enough and that, to the best of my knowledge, should work in sufficiently-standards-conformant ePUB readers. Even if I know it's not perfect (no smallcaps in ADE, slow TOC in Sony, no SVG cover in you-name-it...), I decide to publish it as is, and if someone thinks it's crap, they can blame their reading software and write them hate mail... after all, my book is published for free, with no DRM, and everyone is free to change it
Seriously, though, I think both viewpoints are important, we should care about validation and following specs, and we should also care about real-world performance. Mr. X should have woken up two hours earlier, so he would have had time to make both kinds of check.