In this work an enhanced semi-blind channel estimator is presented for multicarrier CDMA systems with multiple transmit and receive antennas. The novel scheme is based on embedded pilots that are characterized as virtual users and their number is determined in a dynamic fashion according to the propagation channel conditions. The superimposed code-multiplexed pilots are used initially for channel estimation before the despreading operation, thus achieving per subcarrier estimation using the known unique multi-layer pattern of the virtual users. Afterwards, the estimation is enhanced by using the despread pilots. Furthermore, an iterative scheme is incorporated in channel estimation process minimizing the effect of multiuser interference and operates either in interference cancellation, or in virtual user mode. The proposed architecture is evaluated for spatial diversity systems (spacefrequency block coded) with two transmit and up to four receive antennas operating in realistic propagation channels for various modulation and coding schemes. The results demonstrate a significant performance improvement especially as the channel is dominated by its time variant characteristics. The dynamic assignment of spreading codes in the payload and the pilots offers to the system realistic flexibility and adaptivity in demanding channels achieving concurrently maximization of the real overall system throughput, while maintaining low overhead and complexity burden.