As far as I have understood the task is more a merge:
for each file you need to read one line from the number file. Nested loops are not appropriate (and leads to inefficiency).
With shell-builtins

Code:

#!/bin/sh
for a in directory1/*
do
  read i <&3
  {
  echo "${i}.X ${i}.Y ${i}.Z ${i}.T"
  cat "$a"
  } > "${a}withheader"
done 3< file3

The file3 is opened with an extra descriptor 3 where the read reads one line from (could have been just the < and simple read here, but is safer in case some more code would read from stdin).
Like the for-done block gets an opened input file, there is an artificial { block } that gets an opened destination file as stdout.