How large is "large", how many is "many"?

I have the impression that it's rather a space problem, my approach would be to split a hash into a HoH to allow swapping of the second tier.

This works well if you can organize your look-ups in a way (ordering or caching) that accesses to the second tier of hashes is bundled, such that you have a minimal amount of swapping. see also Re: write hash to disk after memory limit and Re: Small Hash a Gateway to Large Hash?

For instance looking up $point{$x}{$y} would be the fastest if you can you can bundle all look-ups to points ($x,*) for one fixed $x.

tl;dr maybe I'm missing the point...(?)

On a side note: are you aware of multi-dim hashes in Perl?

C:\Windows\system32>perl -e "$x=1;$y=2;$h{$x,$y}=4; print $h{$x,$y}"
4
[download]

If yes, why do you you need to join the keys by yourself???

> String: $x . ':' . $y

see perldata#Multi-dimensional-array-emulation