Testing the link latency between nc8q-services and ( w8bi-dara-services and n8jj-services )
with and without tunnel between nc8q and n8jj:

Tunnel from nc8q to n8jj disabled:
gelmce@nc8q-mesh:~$ mtr -w -c 9 n8jj-services
Start: 2022-08-18T06:28:13-0400
HOST: nc8q-mesh                                        Loss%   Snt   Last   Avg  Best  Wrst StDev
  1.|-- localnode.local.mesh                              0.0%     9    0.5   0.5   0.5   0.6   0.0
  2.|-- dtdlink.NC8Q-Kettering-Centerville.local.mesh     0.0%     9    0.8   0.8   0.7   0.9   0.1
  3.|-- NC8Q-Centerville-Kettering.local.mesh             0.0%     9    1.9   2.0   1.9   2.3   0.1
  4.|-- dtdlink.NC8Q-Centerville-HuberHeights.local.mesh  0.0%     9    2.2   2.3   2.1   2.9   0.3
  5.|-- NC8Q-HuberHeights-Centerville.local.mesh          0.0%     9    3.5   3.8   3.4   4.9   0.5
  6.|-- dtdlink.W8BI-DARA-Services.local.mesh             0.0%     9    3.9   4.3   3.5   7.5   1.3
  7.|-- N8JJ-Services.local.mesh                          0.0%     9   43.5  44.3  42.9  48.6   1.9
gelmce@nc8q-mesh:~$
-----
Tunnel from nc8q to n8jj enabled:
gelmce@nc8q-mesh:~$ mtr -w -c 9 n8jj-services
Start: 2022-08-18T06:31:16-0400
HOST: nc8q-mesh                   Loss%   Snt   Last   Avg  Best  Wrst StDev
  1.|-- localnode.local.mesh         0.0%     9    0.5   0.5   0.3   0.6   0.1
  2.|-- dtdlink.NC8Q-hAP.local.mesh  0.0%     9    0.8   0.8   0.7   0.9   0.1
  3.|-- 172.31.187.149               0.0%     9   45.8  44.5  42.1  47.2   1.9
  4.|-- N8JJ-Services.local.mesh     0.0%     9   43.3  44.5  42.3  49.3   2.6
gelmce@nc8q-mesh:~$

Note that the latency between nc8q and n8jj is, more or less, equal via tunnel or 3 RF hops and 3 DtD links.
-----

Testing over a 16 mile link:
Tunnel from nc8q to n8jj enabled:
gelmce@nc8q-mesh:~$ mtr -w -c 9 w8bi-dara-services
Start: 2022-08-18T06:32:34-0400
HOST: nc8q-mesh                        Loss%   Snt   Last   Avg  Best  Wrst StDev
  1.|-- localnode.local.mesh              0.0%     9    0.5   0.5   0.3   0.6   0.1
  2.|-- dtdlink.NC8Q-hAP.local.mesh       0.0%     9    0.7   0.8   0.7   0.9   0.1
  3.|-- 172.31.187.149                    0.0%     9   46.2  45.2  42.0  47.4   2.0
  4.|-- dtdlink.N8JJ-Services.local.mesh  0.0%     9   47.0  45.5  42.6  48.6   2.2
  5.|-- W8BI-DARA-Services.local.mesh     0.0%     9   82.2  83.9  81.4  87.6   2.3

Tunnel from nc8q to n8jj disabled:
gelmce@nc8q-mesh:~$ mtr -w -c 9 w8bi-dara-services
Start: 2022-08-18T06:35:30-0400
HOST: nc8q-mesh                                        Loss%   Snt   Last   Avg  Best  Wrst StDev
  1.|-- localnode.local.mesh                              0.0%     9    0.5   0.5   0.5   0.6   0.0
  2.|-- dtdlink.NC8Q-Kettering-Centerville.local.mesh     0.0%     9    0.8   0.8   0.7   0.9   0.1
  3.|-- NC8Q-Centerville-Kettering.local.mesh             0.0%     9    1.9   2.0   1.9   2.5   0.2
  4.|-- dtdlink.NC8Q-Centerville-HuberHeights.local.mesh  0.0%     9    2.2   2.4   2.2   2.9   0.2
  5.|-- NC8Q-HuberHeights-Centerville.local.mesh          0.0%     9    3.4   3.7   3.3   4.2   0.3
  6.|-- W8BI-DARA-Services.local.mesh                     0.0%     9    6.5   5.0   3.7   6.8   1.3
gelmce@nc8q-mesh:~$

Not that latency increases 16X with 2 tunnels .vs. 3 RF and 2 DtD hops.

IMHO and regardless of complexity,
99% of tunnel.vs.RF routing issues are solvable if the tunnel weight was a user editable setting.

Chuck