This paper presents a detailed experimental comparison of control architectures for delayed bilateral teleoperation. The main goal is to illustrate the key differences in controller performance that can be expected in practice, independent of the human operator, as a function of the communication delay. Existing architectures can be divided into bilateral motion synchronization, where the master and slave controllers implement an as-stiff-as-possible coupling between the master and slave devices, and direct force-reflecting architectures, where the slave controller mimics the operator action, and the master controller reflects the slave-environment interaction forces. Six architectures are analyzed using standard performance indices to assess motion tracking, force reflection, and stiffness reflection quality. In addition, the architectures are also compared on physical operator effort, which is a newly introduced metric to quantify the required operator's effort to execute free motion tasks. The results illustrate that, for increasing delays, direct force-reflecting architectures (in particular, a position/force-force architecture) are superior to bilateral motion synchronizing controllers, in the sense that they are the least sensitive to delays. In contrast, all bilateral motion synchronizing architectures significantly suffer from a reduction in motion tracking or stiffness reflection, or an increased operator effort, when the delay increases. While these conclusions are drawn on a one-degree-of-freedom (DOF) setup, we expect these trends to maintain valid in general, and therefore, the authors suggest that future controller designs for delayed bilateral teleoperation should explore direct force-reflecting architectures more extensively to achieve better performance.