The Ph.D. will always be there, but the opportunity for a startup might disappear with time. Given that, staying upwind (http://paulgraham.com/hs.html) would tend to dictate going with a startup.
Not sure I agree with that; I would suggest that the opposite is true.
Top-tier PhD programs almost never allow students to defer an acceptance offer, so the person in question would need to re-apply. Being accepted becomes substantially more difficult the longer you've been working in industry, particularly if you haven't been working in something directly related to the PhD's field of study (like a startup).
Startup ideas, however, are a dime per dozen, so there is no danger in delaying because it will always be easy to find a worthwhile idea on which to work.
Top-tier PhD programs almost never allow students to defer an acceptance offer
At least in CS, all the top schools I'm aware of usually allow you to defer an acceptance offer (IIRC 6 months for MIT, 1 year for Stanford/Berkeley/etc.).
My startup's based around the tech I was studying for my PhD (I dropped out to do the startup, my research was leaning more real-worldy than academic, and to stay in academia I'd have had to go in a direction I didn't like). I'm pretty sure (in fact I've more or less been told) I could get the PhD within a couple of years without too much hassle, either full time or part time. I agree with another commenter's point that I might not actually enjoy it though!
Computer science is a little different from chemistry, though, it seems easier to jump in and out of academia like that.
Once you've started it's generally OK to take a year off, however if you turn to down an acceptance most places make you start from square one. So as such it might be a good idea to take the PhD position, do that for a year, and then re-consider.
