DOM NodeIterators are computed incrementally, so they do reference the
Document. Xalan's borrowing of that API generally behaves similarly. In
local operation, this can yield a significant performance improvement if
you don't actually use all the retrieved values.
It's been too long since I looked at RMI, so I can't reliably advise you on
how much data will actually cross the wire. Remember too that nodes
themselves reference other data structures, so if the rule is that the
whole reference mesh gets brought with each object shipping the node back
could be equally disasterous.
If in doubt, encapsulate the data retrieval at a higher level, extract only
the information you really need, and make an RMI call to the encapsulation.
