if rxq is buffered it adds only ~50ns to SyncChanRTT (was ~ 350ns ->
400ns) and important: no extra goroutines switches are introduced
(verified via analyzing trace).