Under NPV, the necessary plurality could be confined to a few states, or a single region of the country. Multiple regional or even favorite-son candidacies would be encouraged, and each new candidacy would increase the likelihood of one of them receiving a majority of the electoral votes (courtesy of the NPV compact) while capturing a very low percentage of the overall vote. If there were four major candidates in the race, victory could be achieved with just over 25% of the popular vote.
Hertzberg replies, with more patience than I could muster:
If it were true that popular-vote elections encourage multiple candidacies, we would expect to see multiple candidates in the popular-vote elections we already have, such as gubernatorial and Senate races. We see no such thing.
As for regional candidates, the current system—state by state, with the plurality winner in each state getting 100 per cent of that state’s electoral votes—is actually more hospitable to them, because, by giving such candidates a shot at winning electoral votes, it opens up the possibility that no candidate will win an overall electoral-vote majority, thereby throwing the election into the House of Representatives (where, under the Constitution, each state delegation would get one vote regardless of size). In 1968, George Wallace won 46 electoral votes. In 1948, Strom Thurmond (who got less than two and a half per cent of the national popular vote) won 39 electoral votes. Both were regional candidates, the region being the old Confederacy. (In 1948, by the way, Henry Wallace, a non-regional candidate, got about the same popular-vote percentage as Thurmond but zero electoral votes.)
Sracic is correct that under N.P.V. a candidate could, in theory, win a four-candidate election with a hair over 25 per cent of the popular vote, let’s say 25.1 per cent—but only if each of the other three candidates got exactly 24.96 per cent.
Reading Hertzberg's arguments on this topic, I've come to notice that anti-N.P.V. arguments fall into four categories. You have assumptions that anything the Founders created must ipso-facto be correct. (Bring back the three-fifths rule!) You have arguments based on partisan motive, usually impugning liberals for being bitter about 2000. You have arguments based on lazy misrepresentation of the details of the N.P.V. plan. And fourth, and most common, you have arguments like the one Hertzberg takes apart here, which consist of imagining possible adverse scenarios under an N.P.V. These arguments tend to have certain things in common. They don't assess the likelihood of an adverse scenario, they just invoke it. And they don't balance the possibility of an adverse scenario against the actually existing adverse scenarios arising from the current system.
The last part is really key. Suppose we had a national popular vote, and somebody proposed to change the system to state-by-state winner take all elections. You could raise some really scary scenarios, couldn't you? The less-popular candidate could win! Electors could defect and ignore the voters' instructions! A state legislature could threaten to vote for its favored partisan and ignore the voters completely! Candidates would spend all their time in the dozen closest states and ignore most of the country! Why would we want to set up a method that results in people flying across the country to knock on doors in Ohio or Florida and ignoring their own neighborhood?
That's without even invoking disasters that have not occurred, such as a 269-269 tie, or a candidate who loses the popular vote badly but squeaks into the presidency by concentrating his support in half the states.
Those arguments would be persuasive. Indeed, it would be pretty hard to muster any positive arguments for this reform at all. That's why nobody actually implements it in other countries. And that's why states don't consider getting rid of statewide voting for governor and breaking up the vote into district-by-district winner take all blocs. Because, in other words, the electoral college is terrible.