Eliot,
> can anyone remember the paper (I think OOPSLA) in which the C++ dispatch
> model is inverted and instead of having a dispatch table per class indexed by
> selector id one has a table per selector indexed by class id?
Several such schemes are compared in
Message Dispatch on Pipelined Processors
Karel Driesen, Urs Hölzle, and Jan Vitek
ECOOP '95 (LNCS 952), p. 253 ff
http://www.ifs.uni-linz.ac.at/~ecoop/cd/tocs/t0952.htm-- Jecel