Nepotism alert! It goes to my brother, the terrific reporter David S. Bernstein, who complains about gullible journalists buying the idea that expectation-lowering Mitt Romney might skip Iowa.
As David points out, it wasn't even true that John McCain skipped Iowa in 2008; he shifted resources away, but still was campaigning and competing there. And read him on Rudy Giuliani's 2008 -- excellent. He's correct: Romney is trying to lower expectation in Iowa, not skipping it. Great catch!
I'll add two things. The first, and most obvious one, is that every single serious candidate for the nomination must compete in Iowa and New Hampshire. It isn't necessary to win in both. It might not even be necessary to win one of them, although no modern nominee has ever managed that trick (Updated: see below). But there's just no way to survive the winnowing down effect if a candidate self-winnows in either of the first two states.
The second thing is that, near as I can tell, the idea that one can "skip" these events is purely a throwback to pre-reform days. Before 1972, primaries were mainly a demonstration sport. Formal state party organization officials and state and local politicians, who were the ones who controlled delegations and therefore selected nominees, used information from primaries to help decide if a candidate would "play" in their state in the general election. The most famous example was John Kennedy's victory in West Virginia in 1960, which "proved" that Southerners would be willing to vote for the Catholic Massachusetts Senator. In those days, candidates most certainly did actively enter only selected primaries, and there were appropriate strategies for which primaries -- if any -- to contest. But applying the language of those times to the post-reform world is entirely anachronistic.
I've argued, or at least speculated, that in the current elite-driven process (see Cohen et al.) that primaries have to some extent returned to their old demonstration, signaling function in the process. However, as long as everyone pays attention to Iowa and New Hampshire, and perhaps South Carolina, as the key signals, there's no way any candidate can afford to actually skip any one of them.
See too Josh Putnam's take -- hey, he's quicker than I am!
UPDATE: Dana Houle tweets that neither Bill Clinton nor George McGovern chalked up a win in one of the first two states, and concludes: "Only 14 competitive pres primaries (8D 5R) after 68. Small sample yet in 2 cases nom didn't win NH or IA." He's correct about Clinton and McGovern; Clinton finished second in New Hampshire, while McGovern finished second in both Iowa and New Hampshire. I should have remembered that, above.
However, I don't think it changes much. In 1992, there essentially were no Iowa caucuses; oh, they held them, but not only did the candidates pass on it with Iowa Senator Tom Harkin running, but the press stayed away, thus truly making it a non-event. As far as 1972, the brand-new process was still just getting organized; indeed, no one really understood that Iowa mattered until after the event. Regardless, McGovern and Clinton both contested what there was to contest in the early going. They may or may not be counterexamples of whether or not one needs to win in IA/NH, but they are certainly not counterexamples of the need to compete in those states.