HUGE is clearly the best way to implement this, and this is where the majority of improvements in this area will come from, along with HTTP 1.1.

The file matching does not require exact names -- it uses an approximate matcher to identify files with similar names. I do agree with your point that files of the exact same length for a given query are almost definitely the same file, in which case you could bypass the approximate matching by name all together. We may implement this in our final release.

I disagree that this will signifantly increase the number of swarmable files, however, as it is quite rare that identical files have names that are so different that the approximate matcher does not group them together.

We will likely implement the researching for more hosts for a given download feature at some point.

The best easiest and most robust way of getting more hosts is to build a download mesh -- this is really the ideal, and it is not that hard. We will be doing this fairly soon.

Thanks for the suggestions -- the file length one in particular we may just go ahead and implement.