It seems to me that the three kinds of expressions proposed are
completely orthogonal.  Then why should they all be put in one SRFI?
It forces an implementation to implement all 3 to be able to claim
(cond-expand (srfi-11 'yes)).  An implementor might just implement
case-lambda, and a user might just be interested in case-lambda, but
they can't convey this information to each other through cond-expand.

Marc