Three nitroxide spin probes of different sizes and geometrical shape were used in a 250 GHz ESR study of the probe rotational dynamics in the fragile glass former ortho-terphenyl (OTP) over a wide temperature range from 380 to 180 K. Comparative studies at 9.5 GHz have also been performed. Perdeuterated 2,2’,6,6’-tetramethyl-4-methyl aminopiperidinyl-N-oxide (MOTA), and 3,3’-dimethyloxazolidinyl-N-oxy-2’,3-5 alpha-cholestane (CSL) are, respectively, comparable in size to and larger than the OTP host molecule, whereas Perdeuterated 2,,2’,6,6’-tetramethyl-4-piperidine-N-oxide (PDT) is substantially smaller. The sensitivity of 250 GHz ESR to the details of the rotational tumbling for T greater than or similar to T-c (where T-c is the crossover temperature) was exploited to show that the relaxation is fit by a model that is characteristic of a homogeneous liquid. A nonlinear least-squares analysis shows that below the melting point, T-m, CSL, and MOTA dynamics are well-described by a model of dynamic cage relaxation proposed by Polimeno and Freed wherein the probe relaxation is significantly influenced by a fluctuating potential well coated by the neighboring OTP molecules. A model of simple Brownian reorientation does not fit the experimental spectra of CSL or MOTA as well as the dynamic cage model below T-m. Spectra of PDT do not show any significant non-Brownian dynamics for this probe. It was found that the characteristic rates of the cage model, viz., the reorientation of the probe and the cage relaxation, were describable by activated processes; however, the "average" rotational diffusion rates (defined in the usual manner as the time integral of the correlation function) derived from the dynamic cage parameters follow the Stokes-Einstein-Debye (SED) relation rather well, in agreement with previous studies by other physical techniques. It is then shown that the usual stretched exponential fit to the motional correlation function, interpreted in terms of an inhomogeneous distribution of simple reorientational rates, is clearly inconsistent with the observed ESR spectrum. The absence of a significant cage potential above T-m is discussed in terms of a model of frustration limited domain sizes proposed by Kivelson and co-workers. Evidence for the existence of substantial voids in OTP below T-m, especially from the spectra of the small PDT probe, is discussed in terms of the structure and packing of the OTP solvent.