If you run more than one coding agent at once, what breaks first?

Genuine question, not a survey for anything.

I have been running multiple coding-agent sessions on one machine for a few months. The failure modes turned out to be less dramatic than I expected and more annoying:

Orphaned sessions. A terminal window gets closed, the process does not, and it sits there holding a working directory. I usually find out from load average rather than from noticing.

Two sessions on the same repo. Not the loud git conflict, the quiet one where both are confident the branch is clean.

Port squatting. Something bound 3000 an hour ago, the new session gets a response, and concludes the app it is working on is up.

Disk. Caches, node_modules and build artifacts, times N sessions, until something fails for a reason that has nothing to do with the code.

What I would like to hear from anyone doing the same:

How many concurrent sessions before it stops paying off for you? My own ceiling turned out lower than I would like to admit.

Do you isolate properly with worktrees, containers or separate machines, or do you accept the mess and clean up after?

Which failure actually cost you time, as opposed to the one that looks scariest? For me the expensive one was never a crash, it was a session confidently acting on state that was no longer true.

Asking because I keep building small things to paper over my own version of this, and I would rather find out which of these problems are common than keep optimising for a fleet of one.