That overstatement is measurable, too: naive read is "OCEAN signals" (1 pool = coordinated bloc), but the per-block data says 69% of one pool's templates. If the 11/16 ratio is stable it puts the true committed base nearer 2.6% than 3.8%.
I'll re-run the measurement tonight once 961632 hits and post the before/after — same script, same public API, so anyone can check my numbers against theirs.