This is the 802.11s mesh interface, ie Mode: Mesh Point
See if you can follow my logic:
Batman does its own thing (at layer 3 ipv6) to generate a point to point backhaul connecting the nodes together.
But 802.11s/HWMP generates a multi-point to multi-point network.
In the mpath dump output, a healthy Batman backhaul should only have 1 hop links.
But an HWMP backhaul can have multiple hop links (remember this bit for later!)
The mesh_param dump shows all the 802.11s mesh parameters are at default, meaning HWMP is still active in its simplest mode ( mesh_hwmp_rootmode = 0 ).
Your station dump output shows that all the other nodes are connected to this one, as you would expect from the mpath dump output ie "1" hop.
This means you do not actually have a "mesh" style backhaul, instead it is a "star" type. This is not a problem per se, but .....
I have no idea of your actual physical layout/positioning of the nodes, but looking at the station dump output,
Station 10:7b:44:ce:05:84
is struggling with a very low signal (instantaneous -82dBm, worst case -92dBm), whilst trying to maintain its connection.
The 802.11s HWMP sees this and may well change its connection to one of the nodes closer to it. Batman's layer 3 backhaul routing will try to track this, causing a short dropout, then using the 2 hop layer 2 HWMP route will have greater latency.
Because the HWMP is in its simplest mode, it will, sooner or later, signal permitting, change back to a direct connection (I told you to remember, Batman should only have 1 hop layer 2 links). In turn, this causes another dropout with Batman tracking the change, eventually going back to the poor signal, but lower latency direct path and your pings go back to "normal".
Does this make sense to you ?
A possible fix:
There are lots of things we can do, but the easiest is the following -
On your "main" node, change the mesh_rssi_threshold.
The default setting of 0 (zero) means attempt to connect regardless of signal strength.
Lets try forcing that distant node to connect to a node closer to it, without HWMP pulling the rug.
Currently:
Change this to:
option mesh_rssi_threshold '-65'
Make sure you have committed (saved) the change and reboot.
See what happens. You should probably set the same mesh_rssi_threshold on every node.
If this works, great.
If not, we will need a bigger hammer 
ie start setting some mesh parameters (that cannot be set in a uci config).
