actually if the DHCP does not provide the PXE boot server (bare metal server) IP address through its options configuration (43 and 60 for legacy bios computers, 60 and 67 for UEFI ones), it’s in charge to the PXE server to propose itself as PXE server for the network booting computers.
I’d say that the faster will be the one that will make the target computer to boot from it, so it could be that in different PXE boots the target computer will boot on different PXE servers and, if for example a deployment task is running from one of them and at a following restart, during the task execution, it will boot from the other one, the deployment task will not continue.
So to have deployment tasks to work properly in a subnet with more than one bare metal server, without risking that the target computers boot on different bare metal server at different boots, the DHCP options must be configured properly or, alternatively, you could use a network boot media that specified to the PXE booting target which is the PXE server it must refer to. More details at the following link -> https://www.ibm.com/support/knowledgecenter/SS63NW_9.5.0/com.ibm.bigfix.lifecycle.doc/Lifecycle/OSD_Users_Guide/c_create_dep_media.html
By the way, I think that the first suggestion is, if possible, to have only a PXE boot server serving a subnet.