Well, there's probably three reasons for that:
1) Mason Dixon is not a great pollster in general
2) It was conducted in January before his post NH surge
3) Judging from this poll, where it asks registered Democrats and Republicans who they want to be the nominee, I'm assuming it had the same format in the other poll.
This poll asks Republicans and Democrats only who they want to win, with the results:
Trump 34
Kasich 24
Cruz 23
Clinton 54
Sanders 39
It's difficult to draw too many conclusions on the GOP side considering Rubio, the winner, dropped out. But judging from Trump's awful showing there and Cruz's decent one, it looks like the caucus penalty hit Trump/helped Cruz once again. What effect it had on Rubio is unclear. The poll had him ahead by a couple points in January, but the bigger margin easily could've been explained by him gaining traction nationally as opposed to being caucus specific.
The Democratic side is interesting. The spread is slightly less than her margin among registered Democrats in Michigan, and that's also not taking into account the post NY/post Acela "momentum"/bandwagon jumpers. If Minnesota was a primary rather than a caucus, I'd guess it would have fallen somewhere between Michigan (where she won registered Ds by a similar margin) and Wisconsin (where they tied.) So probably a win for Bernie in the high single digits, whereas she lost the caucus by 24. So the caucus penalty hit her pretty hard as well, as expected.
Conclusion: It's very possible that in January of 2016 Hillary had a 34 point lead among registered Democrats in Minnesota. But they were very foolish, as expected from Mason Dixon, for not including independents in an open primary or more tightly screening for the caucus process. And not conducting a post NH poll just made it even worse looking.