I don't think it's so much an "intel" zero, as the fact that you're probably using floating point math. Not to get into the specifics, but 0.10 actually cannot be exactly represented in "normal" floating point no matter how many bits you use. It's a value very very close, but not exactly 0.10. For floating point comparisons, best practice instead of "value == answer" as a test is to do what you say above, or to test abs(value-answer) < tolerance, where tolerance is something very small.
I skimmed the replies, but didn't see if someone answered this part: the double-equals is a "is equal to" as opposed to a single equals being "gets the value of"--a common usage in programming.
Be warned that the use of text in the conditional (the stuff in quotes) is not dependable. I have been playing with this because I want to display values in base-16, which requires using the letters A-F. I'd have a conditional that displayed text, but it would work in some questions, and not in others, or work for some text, but not for other text (for the exact same condition!).
Last note: In my install, I can do == with the conditional operator, but not <, >, <=, or >= because the > and < characters are translated to > and < respectively by the editor, which is not re-interpreted back as > or < by the calculated question engine. So I end up having to do it the clunky way of checking abs(value1-value2)==(value1-value2) or various other work-arounds. Same for bit-masking -- I can't use value1&1 to find the value of the least significant bit because the & is changed to &
You might be able to tell that I've played with this type of thing a lot. :P
It really be nice to be able to express a format string for a calculated value (like printf), both in the question text and the answers... It's a bit unfortunate that, for calculated questions that require an answer in hexadecimal, I have to restrict the datasets to those that produce hexadecimal results that use only digit values 0-9, and I have to hard-code an occasional A, B, C, D, E, or F into the question text just so that students don't start thinking that hex is another name for decimal.
Yes, I can use a set of questions and pick randomly from that. But that's clunky when I want 30+ versions--like if I need to fix a typo, etc. I also need to separately verify each one in that case instead of verifying the corner cases with calculated and then just adding more datasets. More questions == more places for there to be a problem.