If I understand you — and I very well may not — you'd be generating the whole set and then letting M::C filter it down. Can you illustrate what you're talking about, in Perl?
Caution: Contents may have been coded under pressure.
Comment on Re^4: Efficient Unique Nested Combinations
No I meant to just iterate nested next-combinations but it comes to my attention that actually both loops were degenerate (nCn and nC1) so that M::C would be doing nothing more than returning next elements in arrays, which might as well be handrolled anyway. To assure unique selections from inner arrays, the OPs hash idea seems now as good as any.