Why does (nohup'ed) script need to sleep in order for `ps -ef` to work?

Question

In a script that contains the following lines and is called by ansible (via ssh and then sudo), about 4 out of 5 times, I end up in the if branch, although the process I'm trying to ps and grep is permanently running. The Count: echoed in that case is "0".

PROC=[a]sdf # brackets, so we don't grep the grep process itself
# sleep 0.1
PID=(`ps -ef|grep -E "$PROC"|awk '{print $2}'`)
PIDLEN=`echo "${#PID[@]}"`

if [ "PIDLEN" -ne 1 ]; then
    echo \"$PROC\" does not designate a single, unique process. Count: "$PIDLEN"

Now the funny thing is, when I uncomment the "sleep 0.1" line, I never get into the if branch, which is what I would expect.

This script is started in detached mode using nohup, which may be related. The playbook looks something like this:

- shell:
    cmd: nohup ./script.sh </dev/null >/tmp/out 2>/tmp/err &

---

Update

Irrelevant to my question, but someone in the comments seems to believe that pgrep is always preferable to a ps|grep combo. Just to prove that that's not necessarily the case - it is in fact the reason I do not use pgrep here - try this in debian 10:

$ sleep 1000 $(seq 1 1200)|wc -c&
[1] 13821
$ ps -ef|grep [1]200|wc -l
1
$ pgrep -f 1200|wc -l
0
$ pgrep -f 1038|wc -l
1
$ pgrep -f 1039|wc -l
0
$ pgrep -af 1038|wc -c
4102 # 5-digit PID + SPACE + 4096 chars
$ ps -ef|grep [1]200|wc -c
4952

This shows that pgrep greps/prints only 4096 characters per process, whereas ps -ef does more than that. (Obviously ps|grepping for a number is not a safe thing to do in general, I'm just using this to prove the point.)

All the above is easily verifiable for anyone using this Dockerfile:

FROM debian:10
RUN apt-get clean
RUN apt-get update
RUN apt-get install -y --no-install-recommends procps

which can be built and run using these commands

docker build -t debian10-pgrep-vs-ps .
docker run --rm -it debian10-pgrep-vs-ps