Tomcat thread pools: does the connector's internal pool release idle threads like a shared Executor?

Goal: shrink the memory footprint of a low-traffic Tomcat deployment enough to fit a cheaper hosting tier, by tuning the HTTP Connector's documented thread-pool limits rather than the heap alone.

Two documented mechanisms are in play. The connector exposes maxThreads (long-standing default 200) and minSpareThreads (default 10), and worker threads reserve JVM stack whether or not requests arrive. Separately, a shared <Executor> (StandardThreadExecutor) documents maxIdleTime, default 60000 ms, terminating idle threads above minSpareThreads. The catch: whether the connector-internal pool reaps idle threads after a burst looks implementation- and version-sensitive, while the shared Executor documents that reaping explicitly, so the two are not interchangeable in behavior.

Constraints: peak concurrency must still fit inside maxThreads, and a very low minSpareThreads can add thread-creation latency after a quiet period. Assumption: Tomcat 9 or 10.x with the NIO HTTP/1.1 connector; the config docs shipped with the installed distribution are authoritative for defaults.

Open questions

1. Does the connector-internal pool in Tomcat 9/10 NIO shrink toward minSpareThreads after idle time, or hold its peak size?
2. If idle reaping is only documented for the shared Executor, what semantics change when connectors are moved onto it?
3. What floor for maxThreads and minSpareThreads keeps short bursts from queuing despite low average traffic?