In general, in tunneled mode there is an inner (client-target) and outer (client-proxy) congestion control loop.
In forwarding mode, there is no outer loop.
If the bottleneck(s) are between client and proxy, then the outer loop has a shorter RTT and for most congestion controls will capture more of the bottleneck bandwidth. The throughput advantage of tunneled mode could, in principle, overwhelm the overhead advantages of forwarding mode.
It would be nice to see some simulations or other data to show that this is not a problem in practice before we adopted this work.
In general, in tunneled mode there is an inner (client-target) and outer (client-proxy) congestion control loop.
In forwarding mode, there is no outer loop.
If the bottleneck(s) are between client and proxy, then the outer loop has a shorter RTT and for most congestion controls will capture more of the bottleneck bandwidth. The throughput advantage of tunneled mode could, in principle, overwhelm the overhead advantages of forwarding mode.
It would be nice to see some simulations or other data to show that this is not a problem in practice before we adopted this work.