Incorrect ECMP behavior

[kpfleming]

I've got a 7150 stack running the 09.0.00 'routing' firmware (yes, I know it's not GA, but with 08.x.x I had serious problems with IPv6 routing).

I have OSPF enabled and there are two connected nodes which publish the same destination via OSPF (anycast). Here's the current OSPF route information from the ICX:

      Destination     Mask            Path_Cost Type2_Cost Path_Type192.168.255.1   255.255.255.255 1         0          IntraAdv_Router      Link_State      Dest_Type State      Tag       Flags192.168.120.2   192.168.120.3   Network   Valid      0          3800Paths Out_Port        Next_Hop        Type      State1     lag 1           192.168.101.2   OSPF      0c cb2     lag 2           192.168.102.2   OSPF      00 00

The two nodes advertising this route are at 192.168.101.2 and 192.168.102.2; 192.168.255.1 is used to provide NTP service.

There is a third node at 192.168.66.40 which periodically communicates over UDP to 192.168.255.1. Based on the ECMP documentation, I would expect the traffic from 192.168.66.40 to always land on the same node of the two nodes advertising the route, but it doesn't. Over time, both NTP servers show that they are receiving traffic from 192.168.66.40.

I haven't done anything in the configuration to enable or modify IP load-sharing, since the documentation says it is enabled by default and the maximum path value is 4 (and I only have two routes to the destinations).

Is there something I have overlooked?