The coolest thing about !!Con (other than the amazing talks) is the blinded talk submission process. Talks are purely evaluated on content, not speaker nor the speaker's attributes in any way.
Personally, I think one problem with evaluating talks purely on written proposals is that the process misses out on whether a submitter is able to attract the audience's attention, and to convey their points within the allocated time.
Written proposals can be heavily edited; talks on stage (especially if there isn't a rehearsal beforehand — most conferences wouldn't have the resources to do this) can't.
I admire conferences that rigorously aim for blind selection to give all speaker candidates a fair chance, but I feel that the audience occasionally gets the short end of the stick with a bad presenter.
> So what would work better?
I think a two-pass approach could work better.
The first anonymised pass is for rating and commenting on the written talk proposals.
A second pass, with candidate names revealed, would help fine-tune the lineup and make sure that it's the best it can be. It could also include watching the candidate's talks at other conferences, to see if they're good at communicating their ideas.