Each target receives a cloned payload, runs identical CPU-bound work, and replies with a similar-sized payload. One message in flight per target (round-trip gated, like the project engine). compute is measured inside the target; overhead = round-trip − compute, i.e. the transport cost.
Hypothesis being tested: Chrome ≥127 puts sandboxed (no
allow-same-origin) iframes in their own renderer process
(IsolateSandboxedIframes). A process with no visible frames is
backgrounded; on macOS that maps to background QoS → efficiency cores → slower
compute. Predicted: hidden-iframe compute ≈ 2× worker compute, 1px-iframe
compute ≈ worker compute. If instead overhead dominates, it's the
cross-process postMessage hop. Set work iterations to 0 for a pure transport
(ping-pong) test.
Caveats: Chromium may co-locate the two sandboxed iframes in one
process — for a clean hidden-iframe reading, Reset and run it alone. Confirm
process placement in Chrome's Task Manager (separate “Subframe” rows) or
chrome://process-internals/#web-contents. Cross-check with relaunch
flags: --disable-renderer-backgrounding (priority) and
--disable-features=IsolateSandboxedIframes (process isolation).