As real-time communications are developing rapidly on the Internet, overlay routing has been widely proposed as an easily deployable way to meet their stringent requirements of end-to-end (E2E) performance and interactivity. Contrasting to that much previous work focused on the design of novel protocols, some significant issues impacting overlay routing?s feasibility of performance improvement and guidelines of practical implementation still lack a careful study. This paper contributes to the understanding of three of such technical issues. The first issue is how long it takes a relay node to forward a packet, and what are the determinant factors. The second one is how long the overlay path (OP) can persist to sustainedly outperforming the direct IP-path (DP), and how to select more persistent OPs. The last issue is that given the existence of performance correlation between E2E paths, whether it is possible to gain benefits by cooperatively manipulating multiple paths.