The Licklider Transmission Protocol (LTP) is a standard convergence layer for the Bundle Protocol that is suitable for lossy and long-delay channels. It is part of the Delay-Tolerant Networking architecture that is being developed for space and other challenged networks. In this study, the performance sensitivity of LTP to the choice of the maximum payload size and the signal loss is experimentally verified. This parameter determines how data blocks get segmented and directly affects the total amount of header overhead involved in the data transmissions. While large segments are expected to produce less total overhead than small segments, they entail a larger loss probability which may extend the delivery latency due to the extra sessions needed to handle the segment retransmissions. A series of tests were conducted with the ION-DTN reference implementation of the protocol running over a channel with controllable gain provided by Universal Software Radio Peripherals (USRP), i.e., software-defined radio (SDR). The measurements confirm the tradeoffs involved in the payload size selection and the receive signal level of the channel.