If @in is fairly long, then I'd certainly expect this:

@in = grep { $_->{$field} eq $value } @in;
[download]

To be measurably faster than:

@out = grep { $_->{$field} eq $value } @in;
@in = @out;
[download]

Another optimization you could try is to note that each field you process cuts down the size of @in, so it makes sense to first process the fields that will cut it down the most. Filtering on gender eq "f" is probably going to cut out around 50% of @in, but filtering on surname eq "Davis" is likely to cut out around 99%.

Right now, you're iterating through keys %$where in no particular order - i.e. hash order. Sorting keys %$where to prioritize keys where the data exhibits a high degree of variation ought to speed many searches up significantly, and the only cost will be sorting what is presumably a fairly small list of hash keys.