I don't know about 'optimal', really. The code you list implies that you can rely on having 4-element rows, so this might be some fun:

while (<SRC>) {
   $prod{$1} = [$2,$3,$4] if m{(.+?)\|(.+?)\|(.+?)\|(.+)};
}
[download]

Or, another more general approach with regex captures, which works with any data size:

while (<SRC>) {
   @{ $prod{$1} } = split('\|', $2) if m/^(.+?)\|(.+)/;
}
[download]

Of course, I doubt those are very fast, compared to, say:

while (<SRC>) {
   my @row = split(/\|/, $_);
   $prod{$row[0]} = \@row[1..$#row]; #using a slice;
}
[download]