T1 is a layer 2 protocol, and the T1 extender usually implements a Ethernet-to-T1 encapsulation. The T1-extender would receive the Ethernet frame from the switch, take it as-is (header, payload and trailer) and transmit it byte for byte, using T1 line disciplines. At the other end the reverse happens: The T1-extender would collect all the data out of the T1 signal, re-assemble the Ethernet frame, and send it to the destination switch. Neither the T1 extender, nor the Carrier's equipment, reads or changes any value in the Ethernet frame.
You can view a T1 as a bit-conveyor: You feed data onto the conveyor at a fixed rate, and the T1 delivers the bits in the same order to the other side with minimal delay.
The T1-extender have to implement some control functions, like buffering to account for the discrepancy in speed (frames are received at 10 or 100Mbps, but can only be transmitted at about 1.5Mbps), but not much more.
In my opinion, the trouble can be with the T1-extender. Your carrier should be able to provide performance stats for the end-to-end circuit to prove that the T1 is working between the CSU/DSU's. One possible cause is that the T1-extender receives to many frames, and experience buffer-overflow. To prevent this, you will need some kind of flow-control, that will be dictated by the T1-extender. As far as I know, Ethernet does not natively support a flow-control mechanism on full-duplex links, but the T1 extender might have the option to inject an IEEE 802.3x pause frame to control the flow of packets.
This is all based on theory: I do not have any experience on T1-extenders, and cannot tell you what features typical devices implement. But I hope the info help you to understand the process.