After years of dealing with these kinds of clusterf*cks, I take a different tack. Deadlines that are crazy are the result of poor planning, usually 2-3 levels above me. So, I refuse to cover for their mistakes. I will work hard, and I will make a good-faith effort to make the deadlines, but I will not ruin my life or my family's life for them. Life is too short and there are too many jobs out there that pay well-enough.

I came to this conclusion after I worked over 100 hours in a week (including 18 hours on my wife's birthday) to get a major release out for a company that laid off their entire development staff the next Friday. We were laid off because we asked for something completely unreasonable - time to design. Not only that, but we were allowed to interview for our jobs, but only half of us would get it back. After that experience, I realized that it simply didn't matter.

When I do have to work against a deadline, I follow the "Basic Principles"^TM:

1. Always code the smallest portion that will work at any given time.
2. Build on successes.
3. CVS is my personal savior.
4. Testing, even if it's just basic smoke-testing, is more important than development.
5. Test every single change, one at a time.
6. Have a clean development environment that no-one else can mess with.
7. Spend the time to make the computer do as much of your work for you as possible. The investment will always pay off tenfold in the first week.
8. Don't sweat the small stuff, and it's all small stuff.

Yeah, a lot of those principles sound like XP, but I don't follow the XP paradigm. XP is just a codification of the "Basic Principles"^TM.