Dataset verses Collection

I would like to hear other's opinions on the following topic:

   The distinction between datasets and collections is somewhat artificial.
Currently, a collection is just a container for datasets, while a dataset is
the "atom" that a user can select. The user often makes further selections
from within the dataset.

   The two kinds of selections (within a collection and within a dataset)
can seem very similar, however, selections within the collection are fast
(because the server does not have to be contacted), but for selections
within a dataset, the user can expect more delays because the server is
actually being communicated with. Also selection within a dataset is
protocol-dependent.

   Benno's DODS server seamlessly presents dataset subsets as datasets.
Generally in other DODS servers, the client must use constraint expressions
(CEs) to do this (ie the subset specification is done on the client, not the
server). It would be good to allow catalogers to specify dataset subsets as
datasets when that is supported by the protocol, for example using DODS CEs.

   OK, so two specific questions:

   1) how desirable is it to allow collections to also be datasets, meaning
that they optionally have a URL and can be selected like a dataset, AND they
can be expanded into nested collections and datasets? (Note that this would
not make an arbitrary collection into a datset, but only ones that had a
URL).

   2) If we decide to do 1) then should we simple merge the collection and
dataset elements? My feeling is that it makes things a bit harder to
understand for new users, but is more compact way to implement this.