Context. CoRoT and Kepler missions are now providing high-quality asteroseismic data for a large number of stars. Among intermediate-mass and massive stars, fast rotators are common objects. Taking the rotation effects into account is needed to correctly understand, identify, and interpret the observed oscillation frequencies of these stars. A classical approach is to consider the rotation as a perturbation. Aims. In this paper, we focus on gravity modes, such as those occurring in γ Doradus, slowly pulsating B (SPB), or Be stars. We aim to define the suitability of perturbative methods. Methods. With the two-dimensional oscillation program (TOP), we performed complete computations of gravity modes -including the Coriolis force, the centrifugal distortion, and compressible effects -in 2D distorted polytropic models of stars. We started with the modes = 1, n = 1−14, and = 2−3, n = 1−5, 16−20 of a nonrotating star, and followed these modes by increasing the rotation rate up to 70% of the break-up rotation rate. We then derived perturbative coefficients and determined the domains of validity of the perturbative methods. Results. Second-order perturbative methods are suited to computing low-order, low-degree mode frequencies up to rotation speeds ∼100 km s −1 for typical γ Dor stars or ∼150 km s −1 for B stars. The domains of validity can be extended by a few tens of km s −1 thanks to the third-order terms. For higher order modes, the domains of validity are noticeably reduced. Moreover, perturbative methods are inefficient for modes with frequencies lower than the Coriolis frequency 2Ω. We interpret this failure as a consequence of a modification in the shape of the resonant cavity that is not taken into account in the perturbative approach.