The v1 deep-dive (2026-07-07) measured ~5-10% in-PHP ZTS overhead vs NTS php-fpm. Shrink it:
- Audit ZEND_ENABLE_STATIC_TSRMLS_CACHE=1 is set for php core AND every bundled extension in the spc build (it is confirmed for opcache's config.m4 only).
- Verify PGO is actually active in the gnu lanes (fork is named -pgo- but confirm the profile-use step runs; if not, wire it - typically 5-10% gain).
- Experiment: -ftls-model=initial-exec for libphp.a (statically linked into ephpm, so IE is safe for the main binary; measure interaction with dlopen'd extensions).
Target ships with ephpm v0.4.x (perf release).
The v1 deep-dive (2026-07-07) measured ~5-10% in-PHP ZTS overhead vs NTS php-fpm. Shrink it:
Target ships with ephpm v0.4.x (perf release).