Currently the test will spin up p2p connections in the beginning, then announce the headers to all nodes, but only send the blocks sequentially. This takes a long time, so when getting to the last node, it will have already timed out, while node1 is busy eating blocks. Example:
node2 2022-12-06T19:31:35.419291Z [msghand] [net_processing.cpp:5783] [SendMessages] [net] Requesting block 2cfdb317b3b901f79e2d4f96339d0c0dffd8ef2685d324f62ab0e2fa3402430e (1) peer=0
node2 2022-12-06T19:31:35.424784Z [msghand] [net.cpp:2776] [PushMessage] [net] sending getdata (577 bytes) peer=0
[...]
node2 2022-12-06T19:41:35.423257Z [msghand] [net_processing.cpp:5729] [SendMessages] Timeout downloading block 2cfdb317b3b901f79e2d4f96339d0c0dffd8ef2685d324f62ab0e2fa3402430e from peer=0, disconnecting
node1 2022-12-06T19:41:35.438706Z [msghand] [net_processing.cpp:5783] [SendMessages] [net] Requesting block 6575919043306ed309014d0bd79814b4fab8afaa281e026d8cc3a1c4c2336fbc (1748) peer=0
node2 2022-12-06T19:41:35.521253Z [net] [net.cpp:573] [CloseSocketDisconnect] [net] disconnecting peer=0
node2 2022-12-06T19:41:35.630417Z [net] [net_processing.cpp:1532] [FinalizeNode] [net] Cleared nodestate for peer=0
Fix this by only spinning up the p2p connection right before they are needed.