I believe that These arXiv tags should not be used at all.
My biggest concern is that these tags are not friendly to people who don't know arXiv (I'm one of these people). I have an interest in many computer science topics, but I rarely read computer science publications, have never browsed arVix, and really don't want to learn how to use such a system that I'll never use off of this site.
However, I routinely read other publications of papers by the IEEE (in fact, I have the most recent issue of IEEE Transactions on Software Engineering next to me right now). Papers in this publication have "index terms" - keywords that describe what the paper is about. I grabbed a copy of the ACM Transactions on Software Engineering and Methodology out of my shelf, and they have three sections: categories and subject descriptions (which contain a code as well as textual descriptions), general terms, and additional key words/phrases. Whenever I've searched for papers in an online database, my matches have always come from the keywords sections or the body of the paper.
Looking at these tags, they are unwieldy. I have no idea what the tag means, and I wouldn't expect that most undergraduate computer scientists and software engineers would either. Given the responses to my previous question here on Meta, you want to be inclusive of multiple skill levels. If I find it difficult to search and browse through questions based on tags (which are also used in the site search engine), I'll be less likely to use this site.
Instead of following MathOverflow, I believe that this exchange should follow a path closer to that of Mathematics. There are three relevant discussions on their Meta that I could find. The Mathematics community appears to be doing quite well in terms of having people of various skill levels (appears to range from undergraduate university-level mathematics on up), their tagging appears to be working quite well, and I honestly find it easier to read and search than MathOverflow.
But really, if this Stack Exchange follows MathOverflow, I actually see someone else coming along and making another Computer Science exchange geared toward a broader audience, in terms of skill levels. I think that this exchange can avoid this splintering of the community (avoiding another Unix/Linux versus Ubuntu Stack Exchange problem) by catering to as many people of various skill levels and educational backgrounds as possible. Yes, you need a specific focus on certain topics and clearly define what is and is not on-topic, but you want to segregate based on those topics, not by knowledge or experience. By dumping the arVix style tags, you make it easy for everyone to find questions/answers, but also make the site friendlier to less knowledgeable and experienced people such as myself.
However, that's not to say that arVix isn't useful. I believe Stack Exchanges have the "tag wiki" page that Stack Overflow has. It would be benefitial to identify which tags correspond to arVix topic areas and link to those topic areas in the tag wiki for a given tag.
Also, a technical consideration: The tags are specified using rel="tag" in the source of a page. If you aren't familiar with microformats, search engines actually do pick up on them and they do help people find pages. People who do searches search for things like "data structure" and "data structures", not "cs.DS.data-structures". The tags should reflect what people are searching on Google and Bing so that they are brought to questions on Stack Exchanges that are relevant to their searches.
I also want to address a few comments and other postings.
Another point that I've read in some postings and comments here is that you want to maintain compatibility with MathOverflow. I think this is a mistake. MathOverflow is a Stack Exchange 1.0 site and, last I heard, had no intention of migrating to Stack Exchange 2.0. However, there are a number of Stack Exchange 2.0 sites that might be of interest to this community in some way that are at various levels of existance - Stack Overflow (people might ask algorithm or data structure implementation questions here), Mathematics (which includes discrete mathematics, graph theory, combinatorics...), Statistical Analysis (which includes machine learning), Not Programming Related, Compiler Design, Software Engineering, Artificial Intelligence, Application Security, Machine Learning, Numerical Modeling and Simulation, Signal Processing, Operating systems Development, Natural Language Processing and Computational Linguistics, and Cryptography (and those are just the first 10 pages of proposals, sorted by progress). These are the communities that need to work together to decide what best belongs where, especially since it appears that there will be a feature to move questions between various Stack Exchange 2.0 sites. Unless things change on MathOverflow, you will never be able to migrate a question from here to MO. However, in the future, it will be possible to migrate questions to the exchanges I mentioned, and so far, none of them are using ArVix tagging.
As first mentioned in a comment here, some of the ArVix tags are long. Potentially too long to fit in the limit of a tag. The purpose of tags is to help people search and sort through questions. If you start abbreviating ArVix tags, you open up a whole new can of worms with the best way to abbreviate the tags.