One question per thread please. :-)
a) It depends upon config, but the default behaviour
find gets the primary keys
Using the collection loads a "page" of rows
into a readahead cache from which beans are constructed
as you access the collection element.
The data loaded depends upon your load-groups.
Of course the beans may already be in cache.
Once the original number of rows has been processed,
it loads the next page, etc.
b) Don't you mean CMR? Again it depends upon your config.
The default is to just unlink the data, but you
can specify <cascaded-delete/>