You are raising the perennial question on what model is best. There is unfortunately no clear cut global answer to that: it will depend on the particular needs of your data environment (hard- and software) and not the least the reporting needs that the users express.
Based on those needs you can start to architect a solution for setting up the data: the most common approach is still a star scheme. For teams that are not quite familiar with the effects of queries in a star or snowflake or relation, a star scheme is the best fit, i.e. has the best chance for satisfactory to good results.
When time and budget permits, one can build two solutions and run a few tests to evidence which solution is the best or where the architecture can be changed to improve the performance.
I have no ready title for an international book or article that may compare in more detail the different solutions. The most popular books come from Ralf Kimball & co-authors but he is in favour of star scheme solutions.
One of the authors I know who has a balanced view on this subject is a Dutchman called Rick van der Lans (R20/Consultancy). He has a website on www.R20.nl with an English mirror. You may want to check if he has on his site more info (In his presentations, he has, but the ones I have are in Dutch)