Clearly the removal of old modules is the biggest challenge. It is a well known fact, that a big portion of the software and hardware industry has to cope with "Legacy" CPAN and Perl are no exceptions.
I see several prerequisites for that to be able to happen for every individual module:

• The module must be OIR (old, infrequent, replaceable), that is: last update n years ago, maintainer has no time, not used often, another - better - module exists and there is a migration path. old2new, which the author of the new one provides (as sign of maintenance commitment). That was for old and replaceable. As for "infrequent" I *do*think, that quite raw download statistics do provide a hint. Yes I have read the statement in the FAQ, I have read the lanl.gov website, and i *do* think that this is the sort of pseudo-academic talk that can make every idea seem moldy.
• The module gets a deprecated status for time t. In that time, it either gets improved by the original author, or a new maintainer is found, or a seamless migration path is built (1:1 API in a competitive module).
• After that the module is moved to some CPAN-Nimbus. It is not quite deleted, but if you search on CPAN it is not visible by default.