Too slow readln

[unDEFER via Digitalmars-d-learn]

Hello, there!

I have the next "grep" code:
https://dpaste.dzfl.pl/7b7273f96ab2

And I have the directory to run it:
$ time /home/undefer/MyFiles/Projects/TEST/D/grep "HELLO" .
./strace.log: [pid 18365] write(1, "HELLO\n", 6HELLO

real	1m17.096s
user	0m54.828s
sys	0m13.340s

The same result I get with ldc2..

The same with bash and grep:
$ time for i in `find .`; do file -b "$i" | grep -q text && grep 
-a "HELLO" "$i"; done
[pid 18365] write(1, "HELLO\n", 6HELLO

real	0m42.461s
user	0m23.244s
sys	0m22.300s

Only `file` for all files:
$ time find . -exec file {} + >/dev/null

real	0m15.013s
user	0m14.556s
sys	0m0.436s

Only grep for all files:
$ for i in `find .`; do file -b "$i" | grep -q text && echo "$i"; 
done > LIST1
$ time for i in `cat LIST1`; do grep -a "HELLO" "$i"; done
[pid 18365] write(1, "HELLO\n", 6HELLO

real	0m4.431s
user	0m1.112s
sys	0m3.148s

So 15+4.4 much lesser than 42.46. Why? How "find" so fast can run 
"file" so many times?
And why 42.461s much lesser 1m17.096s?

The second version of grep:
https://dpaste.dzfl.pl/9db5bc2f0a26

$ time /home/undefer/MyFiles/Projects/TEST/D/grep2 "HELLO" `cat 
LIST1`
./strace.log: [pid 18365] write(1, "HELLO\n", 6HELLO

real	0m1.871s
user	0m1.824s
sys	0m0.048s

$ time grep -a "HELLO" `cat LIST1`
./strace.log:[pid 18365] write(1, "HELLO\n", 6HELLO

real	0m0.075s
user	0m0.044s
sys	0m0.028s

The profiler says that readln eats CPU. So why 0m0.075s much 
lesser 0m1.871s?

How to write in D grep not slower than GNU grep?