I am in support of using MO's arXiv-based tags for questions at the advanced undergraduate-early graduate level. The SE platform allows users to filter what they see on the front page by ignoring tags they're not interested in, and I think this will be an absolutely crucial method for increasing satisfaction with the site across multiple types of users who are looking for rather different things.
A big question I see here is how to label questions that are more basic than something that can be given an arXiv classification. There's a basic question on the main page right now about a dice game labeled - inappropriately, I believe - "pr.probability-theory," and I think being loose with those classifications will frustrate users.
The thing about tags based on a highly subjective judgment of level ("high-school," "basic," etc.) is that 1) These tags will most likely have to be adjusted, since the asker will often tag inappropriately, and 2) We've already had accusations of elitism flying around, and however well-intentioned and non-judgmental contributors try to be, there will be those who bristle at level tags, especially when they are necessarily imposed externally in many cases.
Edit: I agree that the use of the two-letter prefix is unnecessary here, and will complicate things for non-arXiv users who still want to ask undergrad-etc. level questions. Some of the arXiv tags won't get used much; perhaps tags should correspond to standard names of subjects as labeled in course names. For example, real-analysis is, I think, a reasonable label, despite not being an arXiv classification explicitly.
I would like, however, to try to make a system where, for example, standard questions about the limit of a sequence or sum of a particular series are not labeled real-analysis: calculus would be appropriate, I think.
As another example, we're going to get a lot of computational matrix algebra problems, so I might advocate that questions about abstract vector spaces be labeled abstract-algebra and not linear-algebra.