The measured and calculated "Rashba polarizations" Deltan/n similar or equal to (0.1 divided by 0.5) in metal-insulator semiconductor (MIS) structures based on semiconductors with small Kane gap \E-g\ (HgCdTe, HgTe, HgMnTe) exceed the ones in the other systems. If we neglect the Rashba splitting, the subband parameters calculated in 8 x 8, 6 x 6, and 2 x 2 Kane models are practically the same, and they are the same for materials with different sign and value of E-g in agreement with experimental data for low-doped samples. In contrast, the Rashba splitting is critically sensitive to taking into account both heavy-hole and spin-off bands. Calculated and measured values of Deltan/n are higher in materials with E-g < 0. The Rashba splitting is essentially nonlinear in the wave vector. This restricts the applicability of the alpha parameter as the measure of the magnitude of the Rashba effect. The linear-in-k approximations can underestimate alpha up to several times. The theory underestimates the Rashba splitting by 20-40%. Allowing for electron-electron interaction reduces but does not eliminate the discrepancies. In narrow- and zero-gap semiconductors, the two-dimensional spectrum, including the Rashba splitting and resonant shift, can be adequately described by the simplest treatment based on a WKB approach.