Dear Mamta,

Here is an explaination from P. Dhrymes book:

"Earlier we saw that the problem of estimating the parameters of a structural

system of equations by 2SLS can be reduced to the problem of estimating by ordinary least squares, the parameters of a "single equation" in the notation of (4.4.23).

However, it was also shown in Section 4.1 that in this context whether such a procedure is efficient or not depends on the covariance structure of the error terms in the various equations of the system. In particular, it was shown that if the error terms between any two equations were correlated, then a gain in efficiency would result by applying Aitken's procedure provided that not all equations contain the same variables. There is an exact parallel in the present case, which would thus lead us to conjecture that, at least in some asymptotic sense, a procedure that would take into account this postulated covariance structure will be efficient relative to the 2SLS procedure, which does not. Of course, it is apparent on intuitive grounds that 2SLS estimation does not use the sample on the entire system efficiently. This is so since, in general, different equations will contain different explanatory variables. If their respective error terms are correlated, then by focusing our attention on one equation at a time we are neglecting the information conveyed by the rest of the system. If we could use such information, then clearly we would improve on 2SLS." (Dhrymes, Econometrics, Springer-Verlag, Berlin, 1974, p. 209)

3SLS is asymptotically efficient.  Both 2SLS and 3SLS provide consistent estimators.