My assumption was, that NNTP articles were close to a drop-in replacement for IB messages.
After reading the RFC, I must admit NNTP's retrieval of articles by traversing a GROUP is less than optimal for throughput and the structure of NNTP is inherently incompatible with the notion of threads, unless the server automatically moved articles to a new "subGROUP".
However I believe we should think about adopting some NNTP concepts such as:
board hierarchies (perhaps with their own overboards)
ability to retrieve resources through WILDMAT
Observing NNTP also gives us the chance to avoid its shortcomings.
Being able to retrieve messages by their ID is undoubtedly a good thing, but you could implement a request for threads or even boards.
Are operations on threads and boards (and/or matching against messages by date) sufficient to negate the cost of linear traversal, or are there more elaborate data structures, that curb the overhead of common usage.