In high performance distributed computing applications, data movements have demanding performance requirements such as reliable and predictable delivery. Predicting the throughput of large transfers is very difficult in paths that are heavily loaded with just a few big flows. This report explores how current high speed transport protocols behave and may improve transfer time predictability of gigabits of data among endpoints in a range of conditions. In a fully controlled long distance 10~Gbps network testbed, we compare several TCP variants behaviour in presence of diverse congestion level and reverse traffic situations. We show that these factors have a very strong impact on transfer time predictability of several transport protocols.