Yes, the inept delay is often set to a value slightly lower than the actual 1/4J value to minimize losses in sensitivity due to relaxation effects. This value should be lower for coherence transfer from 1H to 15N, because the relaxation rate for HxNz(or HyNz) is expected to be lower than that for NxHz (or NyHz) which is present when transferring coherence from 15N to 1H. For actual experimental data on rates of relaxation of the antiphase coherence see the work of Dayie and Wagner.

In principle, you may calculate the optimum time for the inept delays if you know the values of the relaxation rates and the J values. The relaxation rates vary substantially from molecule to molecule and and also depend upon the magnetic field homogeneity as well as on experimental conditions such as pH, temperature, magnetic field strength and viscosity of the solution which in turn depends on the sample concentration. In many cases, even sample concentration changes as a function of time due to precipitation or adsorption on the walls of the NMR tube. In addition, most samples have a small variation in the J values and substantial variation of relaxation rates for different signals. Accurate measurement of the required values HAS to be carried out on the sample of interest only.

Therefore, it is common to use the SIMPLER method of optimizing the INEPT delay experimentally by setting up a 1D experiment in the form of an array with the INEPT delay as a variable (variation of about 10-20% about the expected J value with an array size of about 5-10 is usually adequate) and identifying the delay time resulting in the highest sensitivity.

Note: optimizing the INEPT delay by using a 1D array would optimize the AVERAGE value of all signals. However, in many experiments it is more important to increase the sensitivity for the weaker signals - these are the signals that have high relaxation rates and hence would have lower 'optimal delay time' for coherence transfer step. Therefore, even if the average value of the signals is better by setting the delay time slightly longer than 2.25 ms; for most proteins a shorter delay (~ 2.25 ms) would be best for maximizing the intensity of the weakest signals (because there is a variation in the relaxation rate from residue to residue for most proteins). The logic is that you try to maximize your chances of observing the weakest signals (the less weak signals will be visible anyway, if the weakest are visible/detectable).