I tried to use the netboot feature of the RPi3 and booted via network. Therefore I used dnsmasq of pihole, tftpd-hpa and NFS on another Rpi3. I want to use the RPi headless. The Boot takes up to 30minutes and the ssh logon took more than 5 minutes. I checked the average loads and they had values greater than 10 but the cpu-usage is lower than 20%. The original image is much faster and there is no lack. How to analyze the problem better? I can’t imagine that’s normal.
I've not done PXE boot with an RPi, but with x86 PXE, it is definitely slower than an RPi.
You didn't mention...Is the RPi WiFi or hardwired?
What model exactly do you use from the RPi3 family? Only model 3B+ is optimized for PXE boot. It may be possible also for model 3A+ because it is a stripped down 3B+ but I don't know. So if you use model 3B it may be one reason for your problems.
Then with dnsmasq, pihole, tftpd-hpa and NFS there are some sophisticated applications that must be made to play together. In particular pihole and tftpd-hpa are not needed and complicate things unnecessary. A tftp server is already provided by dnsmasq.
How to setup a proper booting PXE only with dnsmasq and nfs-kernel-server you can find at Netbooting multiple "workers" RPi from a "master" RPi. You may give it at try.