It does not appear to be true that database cost-based optimizers require MANDT, I have added indexes without MANDT in ECC and seen enormous performance benefits using both SQLServer and Oracle DBs.
FWIW I am also OCP certified as an Oracle DBA and have worked as DBA/developer in several other DBs, and I'd be surprised if any CBO post-2000 for an enterprise DB could be so bad at calculating optimization that leaving out MANDT could cause an otherwise high-value index to be ignored.
I tend to believe that using MANDT is generally not a good idea, since production environments tend to be single-client, and even in multiple client situations MANDT cardinality is usually so low relative to the other fields you're using that MANDT would tend to be a wasted hop in the index tree anyway. Of course, in most applications the difference is going to be so trivial as to be undetectable either way (like in Siva kandula's original post), but maybe there's a white paper somewhere.
(I realize this thread is old, but this is a valuable discussion that appears in current search results.)