According to Grothendieck's "Recoltes et Semailles", SGA 5 was totally butchered by Illusie, in a combined effort with Deligne so that it looked useless in comparison with SGA 4 1/2 (wich wasn't a true seminar, and stole some of the missing exposés from SGA 5). This is also the reason why SGA 5 was the last to be published.
Precisely, there were three lost exposés:
- Exposé XI was called "Computation of local terms" or something like that, and was substituted by exposé III-b.
- Exposé IX was about Serre-Swan modules and was published elsewhere by Serre (Linear Representations of Finite Groups).
- Exposé XIII was deleted.
Also, important points remarked by Grothendieck in the oral seminar were suppressed in this edition. It's a real pity, and probably the only way to recover the lost ideas would be to find Grothendieck's original notes (probably the archives at Montpellier University contain them or the IHES library).
EDIT: More about what was missing from SGA 5:
- Exposé IV, about "The cohomology class associated with a cycle", was going to be redacted by Deligne, who instead included it in SGA 4 1/2, chapter 4.
A related theme covered by Grothendieck was "The homology class associated with a cycle". This was discussed throughout many exposés in the seminar, but is absent from the published book. According to ReS, these ideas were published by Verdier in an article with the same name. You can read it here
- Exposé II, as is known from the introduction to SGA 5, was reworked by Deligne and included in SGA 4 1/2 as well.