>I don't mean the function itself is complicated, it is the decision on when and what to round that makes things more complex (do you round to how many decimal places? at the line level or at the totals level? are there different currencies involved in different lines of the same order? how are taxes rounded? are there different kind of taxes with different roundings?)

I've had a case where I had to go for a less acurate procedure. Why? Because the accountants, who were to give the final OK on the whole thing, were calculating so, and they'd want to check my results. For one, I was supposed to calculate the original amount from an amount which contained 12% tax. First I calculated it by

nAmount=round(nOrigAmount*(100/112),2)

...which actually meant multiplying by 0.892857. Which then turned out to be wrong, because they never did it that way - they had the value of 12/112 written down as 10% (and no damn decimals), so my result would differ by 0.007143 - which is a lot, when applied to a sum of few thousand. So I had to rewrite it to be

nAmount=round(nOrigAmount*(1-round(12/112,2),2)

... which is quite wrong, IMO, but it gave them the expected result.

So, it all depends on your users. Accuracy may hurt sometimes :).