#!/usr/bin/perl

use Data::Dumper;
use warnings;
use strict;

die("usage: $0 <key>=<value> [<key>=<value> ...]\n") unless @ARGV;

my %hash;
foreach my $set (@ARGV) {
    my($keys, $value) = split /=/, $set, 2;
    my @keys    = split /\//, $keys;
    my $lastkey = pop @keys;
    my $h       = \%hash;

    foreach my $key (@keys) {
        if (ref $h->{$key} eq 'HASH') { $h = $h->{$key}      }
        else                          { $h = $h->{$key} = {} }
    }

    $h->{$lastkey} = $value;
}

print(
    Data::Dumper
        ->new   ([\%hash])
        ->Useqq (1)
        ->Terse (1)
        ->Indent(1)
        ->Dump
);
[download]

An example usage:

> ./create-nested-hash.pl foo/bar/baz=10 spam/eggs/ham=hi foo/stuff=20
+ top='Tophat!'
{
  "spam" => {
    "eggs" => {
      "ham" => "hi"
    }
  },
  "top" => "Tophat!",
  "foo" => {
    "bar" => {
      "baz" => 10
    },
    "stuff" => 20
  }
}
[download]

While this sort of thing is fun to write, be careful you don't create an unreadable mess. Data-driven programs are an excellent idea; data-driven data structures have a tendency to become unmanageable. The problem is it's far less understandable what the data structure looks like after you go through a few transformations.