Hi,
I guess the answer really depends on your actual use case.
The compound solution is kind of the normal thing to do, and it means that you have a 1-to-1 (or even a 1-to-many) relation between your document and your "group of fields", each instance of the compound is embedded within a document and only available there. If you want to share a certain instance of your group of fields between multiple documents (f.e. a group of category related fields that apply to multiple product documents), you can avoid duplication of content by creating a separate document type to hold your group of fields. Note that a separate document type comes at the price of higher complexity for the CMS user, as you now have to deal with two separate documents, which are edited and published individually.
When you find that the compound is the way to go, and your document needs to contain no more than one instance of the group of fields, you should also weigh the benefits of a compound (code and configuration reuse) against its disadvantages, such as [1], or not being able to use faceted navigation on fields of your field group [2].
hope this helps
Tobi