After doing some looking on this forum I found some info that others may find helpful too:
Particularly helpful from that: POLL_TX_TO_RESP_RX_DLY_UUS + Preamble_length + SFD + PHR + Payload_length < POLL_RX_TO_RESP_TX_DLY_UUS
That helped me understand the relationship between the TX and RX DLY values. Using that, I was able to reduce my DLY values and increase my range accuracy back to the expected ±10cm. Rather than increasing the TX DLY, I was able to increase the initiator RX timeout - this solved all my issues.