Andrew told me the Tb Air team has been considering this, and I'm interested in it as well. Basically, the general idea is to cram as much stuff as possible into the Gloda messages, so that we don't have to stream the whole message through the JS Mime Emitter just because we're missing one specific piece of information. The ultimate goal is to have enough information in the Gloda Message to build a conversation view without streaming the messages, either in Tb Air or in Conversation View. So, this includes having a small snippet.
Hm, I had temporarily forgotten that we already should have the _indexedBodyText available for all in-memory gloda messages. I think what Bryan wanted was snippet summaries available on the conversation object for the two oldest unread messages or something more aligned with display on a conversation scale. In any event, certainly if we could introduce the noun type and stop fetching _indexedBodyText, that could be very useful for performance reasons as it would reduce required disk seeks since we would no longer need to perform the join. Anything else we sketchily pull out of the fulltext table would need to come along for the ride too, though.
After discussion with Andrew, this won't likely happen: i) _indexedBodyText is fine for the conversation, as we'll just cut the N first characters to have a meaningful snippet ii) that would bloat the size of the messages table iii) Bryan is more interested in having the snippets for the very last messages or the first messages, not for every message. So not taking anymore for the time being.