What other than CWND and RWND can cause sending delay?

Investigating some sending latency in a packet capture shows a continual pattern of packet sends sometimes waiting for an incoming Ack:

-> Timely outgoing packet
-> Timely outgoing packet
.. (waiting)
<- Ack
-> Delayed outgoing packet

Strace shows the write calls to the kernal are all timely. TCP no delay is enabled. The two main obvious candidates for the kernel awaiting an ack before sending a buffered packet would be congestion window or receiver window. But there is no packet loss at all, so presumably the cwnd will be sufficiently high, and the window sizes in the incoming acks show no problems with rwnd values. Are there any other reasons that could delay a packet being sent before an ack is received?