3000 polled in NH and only 2% cell-phone ?
Is that most likely a typo ?
I assume they wanted a really large sample in NH for some other reason (polling questions not related to the presidential primaries). I guess landlines are easier to poll, because of a higher response rate? If so, they just didn't need to bother calling that many cell phones in NH, because 2% of such a large sample is already enough that they can just upweight that group to account for any demographic skew. Whereas with the smaller samples, you need to call more cellphones as a fraction of the total number of calls, in order to get a decent demographic sample.
No idea if that's the reason, but it's the most plausible explanation I can come up with.