I don't believe your code is working quite how you think - you're iterating over TMPFILE, then slurping it's contents with the join (which also means you're missing the first line) and then performing a match on that string, repeating the same procedure, and then performing another match on the same string. A simple reduction might look like this (explained below)

open(TMPFILE, "page.html") or die("Can't open page.html: $!")
my($data, $data2);
{
  local $/;
  local $_  = <TMPFILE>;
  ($data)   = /'(\d{3})wordONE/;
  ($data2)  = /(\d{3})wordTWO/;
}
close(TMPFILE);
[download]

That slurps TMPFILE (by undefining $/) into a localised $_ then performs two matches on $_ saving the contents captured in the parentheses into the variables $data and $data2 respectively. This should produce the same (and more reliable) result of the code you demonstrated.
HTH

_________
broquaint