Trying to find where latency in the network happens using ping is, at best, tricky. At worst, not possible. In the best case, you simply have to account for the fact that the router responding to ping is completely unrelated to how fast the router can send a ping packet from one interface to another (most routers respond to ping with close to the least priority, so ping response times tend to vary a lot and are usually much higher than the latency introduced when just forwarding).
The 0.7 ms you have seen can probably be accounted for by the 1905 having more space CPU cycles than the 1700.
Using traceroute (or similar) to find the path a connection takes through the network may be useful, followed by checking traffic levels on intermediate hops. However, that will only reveal delays introduced at routing points. There may be slow-down in switches as well as routers. On the whole, 3 ms is probably not too much of an issue and if it is, there may be other ways of alleviating that. In the past, I've seen (some) OLTP applicatiosn getting slow when FE/BE delay has increased by a few ms, but making sure they can have multiple outstanding queries took care of a lot of that (parallel queries instead of strict serialisation). I'd say it's more common that varying latency usually causes more problem than a constant (although possibly higher) latency.