The majority of the traffic over the Internet is TCP based, which is very sensitive to packet loss and delay. Existing research efforts in TDM-Passive Optical Networks (TDM-PONs) mostly evaluate energy saving and traffic delay performances under different energy saving solutions. However, to the best of our knowledge, how energy saving mechanisms could affect TCP traffic performance in TDM-PONs has hardly been studied. In this paper, by means of our state-of-art OPNET Modular based TDM-PON simulator, we evaluate TCP traffic delay, throughput, and Optical Network Unit (ONU) energy consumption performances in a TDM-PON where energy saving mechanisms are employed in ONUs. Here, we study the performances under commonly used energy saving mechanisms defined in standards for TDM-PONs: cyclic sleep and doze mode. In cyclic sleep mode, we evaluate the performances under two well-known sleep interval length deciding algorithms (i.e. fixed sleep interval (FSI) and exponential sleep interval deciding (ESID)) that an OLT uses to decide sleep interval lengths for an ONU. Findings in this paper put forward the strong relationship among TCP traffic delay, throughput and ONU energy consumption under different sleep interval lengths. Moreover, we reveal that under high TCP traffic, both FSI and ESID will end up showing similar delay, energy and throughput performance. Our findings also show that doze mode can offer better TCP throughput and delay performance at the price of consuming more energy than cyclic sleep mode. In addition, our results provide a glimpse on understanding at what point doze mode becomes futile in improving energy saving of an ONU under TCP traffic. Furthermore, in this paper, we highlight important research issues that should be studied in future research to maximize energy saving in TDM-PONs while meeting traffic Quality of Service requirements.