bjt wilson current mirror
Good explanation by rfsystem on this issue.
I just wanted to add that from experience, other factors to decide between PMOS/NMOS or PNP/NPN for current mirrors are the absolute value of the current and the output impedance
1) For low values of current, say less than 50 uA as a rule of thumb, large channel length PMOS would be better than PNP because of the finite beta (which you compensate for in the PNP design). Also, you add degeneration resistor in the Widlar/Wilson current mirror, so in terms of headroom, PMOS is better than PNP.
2) I know from experience that the output impedance (over frequency) of NPN current mirror in better than NMOS current mirror. I'm assuming the same is the case for PNP, PNP would be better than PMOS (have not verified it, though). The impedance issue is critical if you use it as a tail current in differential pair for high frequency applications.
Based on these factors, I would decide between PNP (NPN) or PMOS (NMOS) for a current mirror in a BiCMOS process.
Bharath