My situation is similar to the above.
My group sees D&D as less "game" and more "story." In that respect, we don't have a good time if the PCs get screwed by the dice and start dying early for no good reason. So I tend to have plausible "threat-reduction scenarios" ready. ;-)
As the campaign progresses, however, I tend not to pull punches. This tends to result in some severely messed-up PCs (e.g. critical injuries, diseases, what-have-you), and maybe one or two dead or dying PCs late in the campaign.
And if the campaign is drawing to a closer or major story transition, then I make a conscious effort *not* to do anything extra to save PCs. Instead I'll make the situation seem as dire as possible. If the PCs are both clever and lucky, its a heroic victory. If the PCs fail and/or some die, it's a heroic tragedy (in the Greek sense), which can be just as good of a story.
My last campaign was 23 sessions and averaged 4-5 players per session. One PC was beyond hope in session 22 and died at the beginning of session 23. A second PC died at the end of session 23 as a result of wounds sustained in the climatic battle.
ironregime