Empirically, IMO the HL technique works best for large producing regions with serious production, about two mbpd or so, for a significant period of time, about two decades or so.  

There are two reasons to do a HL analysis, to estimate recoverable reserves and to develop a model for the world.  

IMO, the two "cleanest" HL models for the world are the Lower 48 and the total North Sea--largely unencumbered production by private companies.  

The Lower 48 peaked at slightly less than 50% in 1970, the North Sea at slightly more than 50% in 1999.   Note that the top 10 major oil companies working the North Sea, using the best data, the best technology, and the best engineers in the world, were predicting that North Sea production would not peak until at least 2010 (Source:  Matt Simmons).  

The total North Sea (crude + condensate) HL plot that I did shows a beautiful linear progression from about 1988 onward.  Qt is 60 Gb, and they are about two-thirds depleted.  

In summary, I think that the premature discoveries of super-giants can cause odd production profiles. So it would be better to factor them out before you do modeling.

No, you should never throw any data away. That is one of the golden rules of the scientific method.  Yes, I know some statisticians like to throw away outliers, but you really should do this only if you understand what the fundamentals are behind the behavior. In reality, what you want to factor out should be part of the underlying model. I might be misunderstanding you, but why not keep the "premature discoveries" in the model?

Take a look at the discovery curves published by Laherrere.

You can see the bimodal components right there. The problem is that the two modes are highly asymmetric in the discovery profile but not so much in the production profile. My own analysis leads me to believe that the second peak gets strongly accentuated by a strong increase in the extraction rate.

I have never believed in the conventional symmetric Hubbert curves, preferring instead to treat the system as a N-order  temporal response to the initial discover stimulus curves.
Ever since about 1995, the extraction rates have progressively climbed so they could keep up with the diminishing returns from the remaining amount.

The full analysis is here

My list of Russian "giant" fields is simply an educated guess, based on various published stories of large fields. Russia's ten largest oilfields account for over 1.8 million barrels a day of Russian production. Its largest producer is still Samotlor, the one Russian field that once produced in excess of 3.5 million barrels of oil each day. Samotlor's output occasionally gets mentioned in various oil journals. The last guess in print had Stamoltor producing 320,000 barrels per day. Almost all the Russian giant oilfields are located in Western Siberia. One giant oilfield outside Siberia is the Romashkino field in the Volga Region. Prior to Western Siberia's oil discoveries, this was Russia's single biggest oil field. It still apparently produces close to 300,000 barrels a day. All of these giant fields are far past peak production. All have high water cuts. Russia is not a low cost oil producer, even in its largest oilfield.

Saudi Arabia Ghawar 1948 4,500
Mexico Cantarell 1976 1,211
Kuwait Burgan 1938 1,200
China Daquig 1959 1,108
Iraq Kirkuk 1927 900
Iraq Rumailia North 1958 700
Saudi Arabia Abqaiq 1940 600
Saudi Arabia Shayba 19751 600
U.S.A. Prudhoe Bay 1968 550
China Shengli 1962 547
Brazil Marlim 1985 530
Iraq Rumailia South 1953 500
Saudi Arabia Safaniyah 1951 500
Saudi Arabia Zuluf 1965 500

Even a cursory look at this list reveals that there have been major changes since the year 2000. It would be interesting if somebody could find an updated list for 2005. I couldn't locate one.

Do you think it is possible to do this analysis for the real fields productions?

Of course obtaining the production for the supergiant fields is a very daunting, even impossible, task but an aproximation, I believe, is plausible