Subtle and hard to trace bugs, yes. Can also happen if time() is used several times in a request, but in different functions. I've taken to preferring $_SERVER['REQUEST_TIME'] in most cases.

If you're spending a few seconds rendering HTML, you probably have other big problems to solve above optimizing around time().

That's the thing. You don't have to. Two subsequent SQL INSERTs may occur only microseconds apart, but the result may still differ by a second, causing a problem with datetime. And it's horribly naieve to think every request always completes in a few milliseconds. There are plenty of good reasons a request may take a longer time. From overloaded servers to network congestion to long running cronjobs or other processes.

Depending on the situation you would want database inserts to reflect the passage of time, and not always stick to a 'start of request/job' time. I can see the desire of the parent to want to have all the HTML generated with the same understanding of time, but as a general practice I still think its a bad one to fix time.