Dovecot stats plugin and mailbox format choice: which metrics justify moving from Maildir to mdbox?

Goal

I want to use Dovecot's built-in statistics support (the stats plugin queried via doveadm) to measure an IMAP performance bottleneck before changing anything, and then use those measurements to decide whether our mailbox format should stay Maildir or move to mdbox. Assume Dovecot 2.3.x on a single host with local storage.

Constraints and uncertainty

The documented optimization paths seem format-dependent: Maildir's fsync-heavy per-message writes and index regeneration show up differently than mdbox's reduced file count and IOPS. Separating index files onto faster local storage (INDEX= path) is also documented, but I do not want to apply it without first confirming that index I/O is actually the bottleneck.

The unresolved decision is that mdbox trades tooling compatibility and simpler recovery for fewer IOPS, and it requires periodic doveadm purge handling. I would like the stats data to drive that trade-off rather than guessing. I am also unsure how much overhead the stats plugin itself adds, which could skew the baseline.

Questions

• Which doveadm stats/metrics counters best distinguish Maildir fsync and index-regeneration bottlenecks from general disk latency?
• At what measured threshold (command latency, I/O wait, session duration) does migrating to mdbox become justified over keeping Maildir with indexes on SSD?
• How should I quantify the stats plugin's own overhead so baseline measurements stay trustworthy?