You're going to have to do two passes if it's a huge file (as in huge = too big for memory). The first pass does most of the work:

• it keeps track of how wide each field is
• writes a temp file with one record per line

The temp file looks something like:

BD3101|bananaswiithapples.gif|\breakfast\fruits\tree\|2007-03-06 14:02
+:31.000000
TP4223|chocolatecaramelfudge.gif|\sweet\desserts\hersheys\|2006-02-28 
+21:16:41.000000
EO2123|tofuwithpeas.gif|\organic\vegetables\legumes\|2007-07-16 13:55:
+06.000000
[download]

The second pass processes this temp file and produces the formatted output:

BD3101 bananaswiithapples.gif    \breakfast\fruits\tree\      2007-03-
+06 14:02:31.000000
TP4223 chocolatecaramelfudge.gif \sweet\desserts\hersheys\    2006-02-
+28 21:16:41.000000
EO2123 tofuwithpeas.gif          \organic\vegetables\legumes\ 2007-07-
+16 13:55:06.000000
[download]

The code that follows doesn't use files at at all (I'll leave that to you - it's trivial) and produces the output above:

my ( @in, @out, @temp_file );
my @lengths = (0) x 4;

pass1();
pass2();

sub pass1 {
    while ( <DATA> ) {
        my @in = unpack "A9A10A9A*", $_;
        if ( $in[0] ) {
            write_to_temp( @out )
                if $out[0];
            @out = @in;
            next;
        }
        $out[$_] .= $in[$_] for 0 .. 3;
    }
    write_to_temp( @out );
}

sub pass2 {
    my $format = join " ", ( map "%-${_}s", @lengths ), "\n";

    for ( @temp_file ) {
        chomp;
        my @f = split /\|/;
        printf $format, @f;
    }
}

sub write_to_temp {
    s/\s+/ /g, s/^\s+//, s/\s+$//
        for $_[3];
    length $_[$_] > $lengths[$_]
        and $lengths[$_] = length $_[$_]
            for 0 .. 3;
    push @temp_file, join( "|", @_ ) . "\n";
}
[download]

PS I've assumed BrowserUk's comment about mistyped sample data to be true.