How should I store the nested view tree?

In Marionette we determine the nested structure dynamically when it is needed. This is done via [this method](https://github.com/marionettejs/backbone.marionette/blob/master/src/view.js#L297-L306).

For no reason in particular I made Puppets work differently. A `_childViews` hash is always available with every view that's nested beneath you.

What are the pros/cons to either approach? At what point would performance be a concern in either situation, if any? Do I _need_ a flat version of the nested tree structure? In what situations would I use it? Or would the dynamically-calculated one be preferable?

Start with the nested-view handling described in src/view.js at lines 297-306, and compare Marionette's dynamic lookup with Puppets' always-available _childViews hash. Document the performance and usage tradeoffs, including whether a flat tree is needed; the issue does not name tests or a specific implementation target.

Written by the indexing model from the issue text.