I've been designing databases and data warehouses in Access, Sybase and Microsoft SQL Server for 12 years, and I say screw the star model. It's fine for small, focused, datamarts, but as you've found out real data is much more complex than that. The "snowflake" schema is better, but I always start off with a fully normalized object-oriented data-modeling approach that provides maximum flexibility for storing and querying data. If your database starts approaching the multi-gigabyte size your queries will slow down, but by that time you will have a good understanding of how your data is being used and you can start pre-aggregating data or generating smaller datamarts from subsets of the data.