On the basis of the code you have posted, you are making really hard work of parsing those files.

Commensurate with refining the templates for records 1 & 2 which you haven't fully described, something like this would read the entire thing:

#! perl -slw
use strict;
use constant {
    RECORD_1    => 'f512 l512',
    RECORD_2    => 'f3 l3',
    RECORD_3    => 'ff ll s2',
## index of field in rec2 that contains count of type 3 records that f
+ollow it
    COUNT       => 3, 
};

open I, '<:raw', $ARGV[0] or die "$ARGV[0] { $!";

my @rec1 = unpack RECORD_1, do{ local $/ = \4096; <I> };

until( eof( I ) ) {
    my @rec2 = unpack RECORD_2, do{ local $/ = \36; <I> };
    for ( 1 .. $rec2[ COUNT ] ) {
        my @rec3 = unpack RECORD_3, do{ local $/ = \28; <I> };
    }
}

close I;
[download]

Note also, that on recent versions of Perl (since 5.10), unpack can deal with little/big-endian issues for you.

Say your type 2 records contain 2 big-endian floats; followed by 2 little-endian int32s and then 2 big-endian int32s:

Use a template of:'f>2 l<2 l>2'