Magnus idea was to first optimize the collector to make it less of a problem to collect more data. Sounds like a good thing to do, but maybe more data in it wouldn't be a problem as long as you don't read too often from it?
The stats collector is a known problem under certain circumstances, so improving it would probably be a good thing. The first thing that comes to mind is splitting it into more files.
Is there any chance the project would accept a patch which adds the pg_stat_*_columns-feature without first optimizing the collector? I guess it primarily depends on how much of the new code that would need to be rewritten, if the collector is optimized/rewritten in the future?
(I would be interested in sponsoring the work, if anyone is interested.)