This is one of the strange behaviours of the MX. I have never seen any documentation on this, but based on observations I would describe it in the following way (to compare it to classical IOS routers):
- WAN1 has its own VRF with only a default route and it's own SLA-config to check internet-reachability
- WAN2 has its own VRF, similar to WAN1.
- All internal VLANs, static routes and also learned VPN-routes share a common VRF
- Based on the reachability-checks and the flow-preferences there is some route-leakage and inter-VRF NAT to direct the traffic to WAN1 or WAN2
Something like this could be the reason that the directly connected networks are not always reachable.
If you found this post helpful, please give it Kudos. If my answer solves your problem, please click Accept as Solution so others can benefit from it.