Just did some playing around with this idea.
The problem lies in that Coda is unable to show the same row twice in a view. So to achieve this:
You really have to have two rows for item 1, two rows for item 3 and so on.
In short, one row for each combination movie-tag. So in theory the best solution is to have a big table for all combinations and then just make views from it. The caveat here is you lose many features like drag-and-dropping or using select lists.
I was able to create a semi-decent workflow to keep these features by using two separate tables, then running actions to sync the data. This is far from best practice, I still vouch for @justin 's method.
More explained inside the doc: