We are getting to the point with schema where we can load a node with just a single query. It needs bit more work, but i think this is a solvable problem for D7.
Really depends on what we to achieve. If we say, one query to retrieve the nids of the nodecomments and then one query to load a node, then probably yes. You do not want to JOIN all this into one huge query, see the one-to-one tables considered harmful on why (keys can't span tables, filesort hell). And adding foreign keys might happen in D6, it's not a big deal :) So at best we have one more query per nodecomment compared to what we have now. That's better than the five or so I feared originally but still :)
Some versions of the node_rendering patch had a 'style' whereby nodeapi 'view' op was skipped. That might be useful here. Otherwise, I think we need to look at each slow nodeapi op and accelerate it, or make it configurable by type. I am encouraged that Robert's sites are working fine with nodecomment.
I am not. Look at http://groups.drupal.org/node/3550#comment-10454 -- there are few nodecomments per node. And whatever you do, it'll be slower than just slapping the filtered comment body to output. However, I am not too afraid of this part -- we can overcome these problems. So it seems the next step is adding foreign keys to schema and bench the outcome. Regards, NK