Bitmasking does in many ways break relational integrity, as you end up storing multiple distinct values in one column, as opposed to one value per column. (It's like Joe says: One fact one way one place one time.) However, it is very efficient at storing quantities of binary data, and coding for one binary(9) column is vastly simpler than coding for 500 bit columns, and that makes it very tempting. I implemented such a solution once, and regretted it within the month as it became very awkward to analyze the state/setting of one value in my relatively small set of 10.
I think that it depends upon the data being represented. If you had a very tightly coupled set of binary values, such that together they could reasonably and practically be considered a single data entity, then it might be acceptable and reasonable to use this tactic. Otherwise, beware the long-term maintenance costs and data obfuscation! (It may well be that, as with with star schemas, the value returned by this kind of design outweighs the costs of breaking conventional relational architecture. It's important to understand and appreciate the trade-offs you're making when you design systems.)