Some comments based on the slides:

• Page 8 section 1 should mention that locked hashes provide a good solution to this problem. But are restricted to post Perl 5.8
• The point on page 16 about refaddr having a cost to calculate is IMO a bit of a red-herring. Returning the refaddr of a SV doesn't really involve any calculation, simply the creation of a new SV and the copying of the relevent address. I would guess that this is roughly equivelent to dereferencing a ref to a scalar in terms of processing. Hashes are going to be slower however because the conversion will be something like void * => newSvIV => newSvPV (actually I think its a SvUpgrade or whatever :-). But this isnt refaddr's fault. For instance it wouldnt be a too difficult to put together an XS routine that did all of this without the intermediate steps. At some point I bet it would be competitive with the scalar index dereferencing.

Otherwise id say your slides are great and are well worth reviewing for a good example of how to do a good presentation on a subject. Great work++.