I never really liked those star and snowflake schemas, they seemed to be building too much of the supposed application into the data model. To me, a data warehouse should be as neutral as possible to support future possible applications. I’ve always preferred that the data warehouse be close to the OLTP source and maybe even more normalized to remove the OTLP performance hacks. We understand the OLTP data model because it supports the real business, if the data warehouse is more or less the same, it is also easy to understand. Those star schemas just confuse me, I can’t tell if the data is in any way correct. If you need a star schema as a hack for a particular problem, it is fine to roll one up as a copy (or data mart) but don’t build the warehouse on that sort of thing.
Column oriented databases are interesting. See <http://www.databasecolumn.com/>. The fact that the legendary Michael Stonebraker is involved in this alone makes it interesting. I think they’re onto something and it seems to be a way to get around some of the performance issues without the complexity of star schemas.